10^th DeKUT International Conference on Science, Technology, Innovation and Entrepreneurship

1.Development of a Low-Cost Atomic Force Microscope Using a DVD Optical Pickup UnitAbstract

Macrey Bwalei^1*⁠iD, Ephraim Mwangi Kinyua^1*, Esther Clara Auma¹, Valerie Olali¹, Gloria Chepchumba¹, Wamburu Gakinya¹, Luke Otieno¹⁠iD

The atomic force microscope (AFM) is a critical tool for measuring forces and characterising surfaces at the nanoscale, with applications spanning biology, chemistry, materials science, and micro-/nano-electronics. Commercial AFM systems typically range from tens of thousands to over half a million US dollars, placing them out of reach for researchers in poorly resourced institutions and developing countries; for basic models, this cost is largely driven by specialised opto-electro-mechanical components and electronic controllers. Building on prior open-hardware efforts to repurpose optical storage drives for nanoscale sensing, this project develops a low-cost AFM by leveraging salvaged opto-electronics from a consumer DVD drive as the cantilever-deflection sensing element, with some of the instrument's opto-mechanical parts produced by 3D printing. Custom electronics for optical focusing will be developed, together with data acquisition software for real-time monitoring and surface visualisation. These components will be integrated with a commercial cantilever and lateral nano-positioner, and the system's performance will be validated by imaging the tracks of an optical disk of known geometry and comparing the resulting resolution against reference values. The resulting instrument is intended to provide an accessible nanoscale characterisation tool for the local research community, supporting advances in materials research and other nanoscale imaging and manipulation applications.

Keywords: atomic force microscopy, optical pickup unit, nanoscale displacement

DOI https://doi.org/10.33803/dg4tqzPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/dg4tqz ↗

2.From Insects to Sensors: Bio-Inspired Approaches to Functional IntegrationAbstract

Kiril Dimitrov^1*⁠iD

The development of sustainable functional materials is an important step toward resource-efficient and smart food production. Insects represent a promising renewable source of chitin, which can be further converted into chitosan and processed into materials with versatile mechanical, chemical, and functional properties. This work focuses on the development of insect-derived chitin and chitosan as functional materials for piezoelectric sensing elements and their subsequent integration into composite structures. The investigated material pathway begins with the extraction and purification of chitin from insect biomass, followed by its controlled conversion into chitosan. Particular attention is given to the influence of extraction, deacetylation, and processing conditions on the resulting material properties. Chitin and chitosan are subsequently processed into films and composite structures and investigated regarding their morphology, mechanical behavior, stability, and electromechanical response. A central objective is the development of bio-based materials capable of generating an electrical response under mechanical deformation, pressure, or vibration. The intrinsic properties of chitin and chitosan, as well as their combination with additional functional phases, provide opportunities to tailor the piezoelectric performance and mechanical characteristics of the resulting materials. The developed sensing materials are further considered for incorporation into composite structures, where mechanical and sensing functions can be combined within a single material system.

Keywords: insect-derived chitosan, piezoelectric materials, functional integration

DOI https://doi.org/10.33803/tp22q3Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/tp22q3 ↗

3.AgriVery: Decentralized Smart Solar Drying and Quality VerificationPoster

Antony Mulwa Waruiru^1*

Background. Small-scale farmers face significant revenue destruction due to inefficient open sun-drying and predatory supply chains. This project set out to design a decentralized, automated solar drying chamber to standardize crop quality and provide farmers with direct access to premium markets without middlemen.

Methods. The system was developed as a mechatronics engineering project at Dedan Kimathi University of Technology. We designed a 3m x 2m x 2m hybrid solar chamber utilizing passive thermal storage and an active environmental control loop. Industrial temperature and humidity sensors fed data to a microcontroller to dynamically adjust exhaust fans, measuring crop moisture reduction over three days.

Results. Testing demonstrated that the AgriVery hybrid chamber significantly accelerated the drying process compared to traditional methods. Over a 3-day period, the mechatronics system reduced crop moisture content from an initial 80% down to the critical 12% safe storage threshold, whereas traditional open sun-drying only reached 35% moisture in the same timeframe.

Conclusions. This automated approach transforms smallholder farmers from vulnerable raw material suppliers into verified quality producers. By standardizing post-harvest processing at the village level and logging yields into a cloud inventory, the system allows agricultural communities to bypass brokers, capture higher retail margins, and scale operations rapidly with minimal infrastructure overhaul.

Keywords: mechatronics, solar drying, post-harvest processing, smart agriculture, quality verification, cloud inventory

DOI https://doi.org/10.33803/tfm784Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/tfm784 ↗

4.Digital Financial Inclusion, Climate Resilience and Food Security in East Africa: Does Inequality Shape the Digital Finance–Agriculture Nexus?Abstract

Dennis Kinini^1*

Food insecurity is a long-standing development problem in East Africa, where farming systems are becoming more and more exposed to climate change and access to financial resources and socioeconomic inequalities. Meanwhile, the rapid expansion of digital financial services offers new opportunities for households and agricultural actors to access payments, savings, credit and risk-management services. It is, however, not yet well understood whether digital financial inclusion and climate resilience enhance food security, and whether inequality conditions the effect. This study explores how digital financial inclusion and climate resilience influence East African food security and the moderating effect of income inequality. The study is based on secondary panel data for East African countries and the two-step System Generalized Method of Moments (SGMM) estimator is used to deal with dynamic persistence, unobserved country heterogeneity and endogeneity. The findings suggest that digital financial inclusion, climate resilience are statistically significant and have a positive impact on food security, while income inequality is statistically significant and has a negative impact. Next, the relationship between digital financial inclusion and income inequality is negative and statistically significant, meaning that increasing inequality reduces the effect of digital financial inclusion on food security. The results suggest that simply increasing digital financial services might not be enough to enhance food security when there are financial barriers to entry for vulnerable groups. The study thus calls for policies to address inequalities and constraints in access to digital financial services and to build climate resilient agriculture to achieve inclusive digital financial services for improving food security in East Africa.

Keywords: digital financial inclusion, climate resilience, food security, income inequality, east Africa, system GMM

DOI https://doi.org/10.33803/9m6vtqPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/9m6vtq ↗

5.Joint Optimization of Screening Thresholds and Inspection Intervals for Edge Assisted Kale Disease SurveillanceAbstract

Michael Muigai^1*⁠iD, Peterson Nyaga¹, Morris Kimathi Mwangi¹⁠iD, Samuel Macharia Kang'iri^1*⁠iD

Edge-assisted crop disease monitoring fixes the capture schedule by convenience and tunes the offloading threshold for accuracy. They are coupled through the epidemic they are meant to catch: a leaf cleared in error costs whatever the disease gains before the next inspection, so we solve them jointly. An ESP32-CAM node computes a chlorosis and necrosis index from each kale leaf image and reaches a MobileNetV2 classifier only over 2G/GPRS. Class conditional densities of the index make screening a likelihood ratio test whose threshold fixes offload probability and edge miss rate in closed form. A renewal model of the uplink, parameterized by measured goodput and geometric retransmissions, prices an offload in seconds and joules; goodput, far below the nominal 85.6 kbps, sets what each image costs rather than how many are sent, which is how link quality enters the optimization. Late detection is costed by van der Plank's logistic model, standard for polycyclic foliar epidemics, whose apparent infection rate is swept over the 0.10 to 0.45 per day range reported for brassica, not fitted. The objective is an expected daily cost weighting crop severity lost to delay, node energy, and transmitted bytes, minimized over threshold and interval by grid search with local refinement. At a 2% edge miss rate, the optimum offloads 18% of captures, cutting transmitted volume by 81% against offloading everything and adding 1.3 days of delay at a cloud recall of 0.92. Link quality leaves the threshold unmoved but stretches the interval from 8 to 16 days as energy rather than airtime becomes scarce; optimizing jointly is worth a few percent. Goodput, retransmission rate, transmit current, and offload latency are measured on the node; energy, delay, and risk are model outputs. Measured latency matches prediction within 12% at a cloud macro F1 of 0.92.

Keywords: hierarchical inference offloading, plant disease epidemic modeling, detection threshold optimization, bandwidth constrained IoT, embedded computer vision

DOI https://doi.org/10.33803/yrtpzmPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/yrtpzm ↗

6.Nematode Community Profiling in Organic and Conventional Spinach SystemsAbstract

Scolasticah Munene^1*, Hannah Karuri^1*

Agricultural management practices play a significant role in shaping soil biological communities and ecosystem processes. Soil nematodes are valuable biological indicators of soil health because they occupy different trophic levels and are sensitive to alterations in environmental conditions. This study assessed the influence of organic and conventional farming systems on soil nematode communities in spinach production systems in Embu County, Kenya. We hypothesized that organic farming systems would support higher nematode diversity and a more complex community structure than conventional systems. Soil samples were collected from farms under organic and conventional systems and analyzed using high-throughput sequencing targeting the 18S rRNA gene. A total of 394 amplicon sequence variants (ASVs) representing 73 nematode genera were identified across the two systems. Differences in the relative sequence abundance of nematode families, genera, and trophic groups were observed between the farming systems. Cephalobidae, Prismatolaimidae, and Trichodoridae showed higher relative abundance in the conventional system, while Monhysteridae and Qudsianematidae showed higher relative abundance in the organic system, but these differences were not statistically significant. Acrobeloides, Prismatolaimus, and Paratrichodorus showed higher relative abundance in the conventional system, whereas Eumonhystera, Diploscapter, and Microdorylaimus showed higher relative abundance in organic farms; these differences were not significant (p > 0.05). Similarly, fungivores and plant-parasitic nematodes showed higher relative abundance in conventional farms, whereas bacterivores, predators, and omnivores showed higher relative abundance in organic farms, although these differences were also not statistically significant (p > 0.05). No significant differences were observed in alpha or beta diversity between the two farming systems. Our work provides a valuable baseline for future studies aimed at enhancing sustainable management of spinach agroecosystems.

Keywords: organic amendments, soil biodiversity, soil health, sustainable agriculture

DOI https://doi.org/10.33803/ed3aqtPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ed3aqt ↗

7.Physico-chemical Sensory and Storage Stability of Fermented Purple TEA Beverage Sweetened with Yacon SyrupAbstract

Dinah Ochami^1*⁠iD

Kombucha is a traditional fermented tea beverage that has gained popularity as a functional beverage because of its antioxidant and probiotic properties. However, there is minimal research on using local plant ingredients to maximize its organoleptic and nutritional qualities. This study examined the formulation of innovative kombucha using four formulations of Kenyan purple tea enhanced with different amounts of yacon syrup as a sweetener: Treatment A (control, 100% purple tea with 5% sugar), Treatment B (95% purple tea, 5% yacon syrup), Treatment C (90% purple tea, 10% yacon syrup), and Treatment D (80% purple tea, 15% yacon syrup). Following AOAC guidelines, samples were examined for pH, titratable acidity, soluble solids (°Bx), and alcohol concentration. The Folin-Ciocalteu procedure was applied for measuring polyphenols, the DPPH radical scavenging assay was employed for assessing antioxidant properties. Mineral content was measured using Atomic Absorption Spectrophotometry. Trained panelists assessed sensory profiles on a 5 point affective hedonics scale of each treatment's aroma, taste, color, flavor, astringency, mouthfeel, and acceptance. Significant increases (p < .05) in physical measurements (e.g., pH = 4.1-4.7; titratable acidity = .46-.59%; soluble solids = 6.3-8.3 ° Bx; alcohol % = 0.5-1.2%) were found across treatments with yacon syrup compared to controls. Greater levels of yacon resulted in enhanced antioxidant and polyphenolic values. Yacon syrups significantly increased minerals including calcium, magnesium, zinc, and iron. In the sensory evaluation, the combination of 90% purple tea & 10% yacon syrup ranked first on all variables. Results indicated that formulations of yacon into purple tea improved flavor; lowered astringency, improved mouthfeel and improved total acceptability. Evaluations of stability demonstrated that all combinations remained consistent in pH and chemical stability when stored under two different conditions (ambient temperature 25°C and refrigeration 4°C verifying the stability of the product.

Keywords: polyphenols, fermentation, probiotics, prebiotics, antioxidant, functional beverage, minerals

DOI https://doi.org/10.33803/9a7qz2Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/9a7qz2 ↗

8.Physicochemical, Bioactive, and Sensory Properties of Hybrid Solar-dried Soursop (Annona Muricata) Whole Fruit LeathersPoster

Daniel Gakuha Wahome^1*⁠iD, Daniel Mwangi Njoroge¹, Amos Kipkemoi Ronoh¹

Background. Soursop (Annona muricata) is a nutrient- and bioactive-rich tropical fruit that is highly perishable due to its high moisture content. Fruit leather is a dehydrated fruit product. Soursop peels and seeds are often discarded as processing wastes, whereas they have been reported to contain higher levels of nutrients and bioactive compounds than the pulp. This study therefore evaluated the effect of substituting soursop pulp with seed and peel powders on the physicochemical, bioactive, and sensory properties of hybrid solar-dried fruit leather.

Methods. Four formulations were made: control (100% pulp), T1 (90:5:5), T2 (80:10:10), and T3 (70:15:15) peel and seed powders, and dried at 55–62 °C to a moisture content of 12.44–13.11%. An affective test was used to establish consumer acceptability. Data was analyzed using one-way analysis of variance, and means were separated using Tukey's test at p < 0.05.

Results. Substituting soursop pulp with peel and seed powders increased crude protein from 1.77- 3.16 g/100 g, crude fiber 1.70-5.43 g/100 g, and crude fat 0.64 -1.94 g/100 g. The carbohydrate content decreased from 82.45-76.89%. Total phenolic and total flavonoid content increased from 111.13-128.53 mg GAE/100g, and 78.63-99.93 mg QE/100g, respectively. Antioxidant activity increased as indicated by a reduced 2,2-diphenyl-1-picrylhydrazyl IC₅₀ from 202.08-121.11 µg/mL. Overall likability remained acceptable, with the leather containing the lowest pulp recording the lowest scores.

Conclusions. This study demonstrates the potential of utilizing soursop peel and seeds in the production of value-added functional fruit product with enhanced nutritional and bioactive properties. Additionally, utilization of these commonly discarded fruit by-products can help reduce processing waste, promote whole-fruit utilization, and support a more sustainable and circular fruit-processing system.

Keywords: Annona muricata, fruit leather, functional, fruit by-products, valorizing

DOI https://doi.org/10.33803/jfkzayPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/jfkzay ↗

9.Physicochemical, Consumer Acceptability and Stability of Passion Fruit Juice Supplemented with SeaweedAbstract

Charity Mutheu Kimeu^1*⁠iD, Eddy Owaga², Daniel Mwangi Njoroge³

Passion fruit juice is very prone to phase separation in storage and should be stabilized by the addition of stabilizers. In this study, the physicochemical properties, bioactive components, mineral profile, consumer acceptance and stability of passion fruit juice stabilized with Kappaphycus alvarezii seaweed gel were investigated. Seaweed gel was incorporated at 0% (JSW0), 5%(JSW5), 10%(JSW10), and 15%(JSW15). The addition of seaweed had a significant effect on various juice parameters (p < 0.05). Syneresis significantly reduced from 66.78% in JSW0 to 16.53% at JSW15 supplementation whereas viscosity increased significantly from 5.22 cP in JSW0 to 24.22 cP in JSW15. Seaweed supplementation significantly increased the antioxidant capacity, as seen from the DPPH radical scavenging (44.35 ± 0.2 µM TE/mL) and FRAP (80.24 ± 0.2 µg FCE/mL) in JSW15. Furthermore, seaweed supplementation markedly increased mineral content, including sodium (125.23 mg/100 g), potassium (123.12 mg/100 g), calcium (172.64 mg/100 g), magnesium (70.15 mg/100 g), iron (40.05 mg/100 g), manganese (0.30 mg/100 g), and zinc (4.01 mg/100 g) in JSW15. The sensory evaluation results revealed that JSW5 had the highest scores for appearance (8.66), color (8.80), and overall quality (8.40), and the highest purchase intent (4.97). As the concentration of seaweed increased (JSW10 and JSW15), sensory rejection was observed due to excessively thick consistency, brownish discoloration, and a lingering, savory marine aftertaste with increased seaweed concentration. Seaweed decreased the syneresis of the passion fruit juice by preventing phase separation during storage for 21 days at 4°C. Hence, 5% supplementation is ideal for a perfect balance of functional values with consumers' acceptance. The results indicate that seaweed gel is a suitable natural stabilizer and fortifying agent for passion fruit juice.

Keywords: passion fruit juice, seaweed gel, antioxidant capacity, mineral content, natural stabilizer

DOI https://doi.org/10.33803/r2jmkbPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/r2jmkb ↗

10.Popularizing African Indigenous Vegetables in Kenya: Traditional Foods to SuperfoodsAbstract

Peter Ogalo Ogweyo^1*⁠iD, Peninah Wanjiku Chege^1*, Faith Nungari Wanjiku^1*, Carolyn Jerotich Koech^1*

African indigenous vegetables (AIVs) are rich in nutrients, yet they are used less than expected in Kenya. Even so, they can support better food choices, stronger food systems, steadier income, and farming that puts less strain on land. This study examines what makes AIVs harder to adopt and what could help them spread, guided by five research questions: (1) What are the production and consumption patterns of AIVs in Kenya, and how do they differ between rural and urban households? (2) What is the evidence linking AIVs to nutrition, food security, and dietary diversity? (3) What are the nature, scale, and causes of postharvest losses in AIV supply chains, and what processing or value-addition options exist? (4) How do consumers perceive AIVs and value-added products, and what shapes their willingness to pay? (5) What policy, institutional, and hospitality-sector arrangements are needed to move AIVs from traditional foods to market-ready products?The work uses a desk review, drawing from published studies on consumption, production, harvest handling, postharvest practices, processing, and buyer preferences. Evidence shows a clear gap between rural and town consumers: rural households eat AIVs about twice as often as urban households. Postharvest losses may reach 50% or higher. Intervention evidence shows that combining nutrition education, cooking changes, and farming steps can improve diet variety, with one program reporting a women's dietary diversity score of 5.1 versus 4.2 in the control group (p = .035). From these findings, the study proposes a research-to-market pathway linking nutrition proof, processing methods, cooking trials, branding, institutional buying, and consumer demand. The review urges Kenya to shift from a production-only focus to a system that starts with how people eat, with more room for culinary innovation and new product ideas.

Keywords: African indigenous vegetables, food security, nutrition, value addition and culinary innovation

DOI https://doi.org/10.33803/7gp7mnPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/7gp7mn ↗

11.A Self-adaptive Model for Jigger Infestation Detection and Prevention in KenyaAbstract

Hillary Kiplagat Benjamin^1*

Jigger infestation, caused by Tunga penetrans, remains a public health challenge in Kenya, particularly in rural and underserved communities where poor sanitation, limited footwear, and restricted access to healthcare increase vulnerability. This study developed a self-adaptive model for early detection, risk assessment, and prevention of jigger infestation by integrating image-based machine learning with structured household, clinical, behavioural, and environmental information. The study was guided by four objectives: to identify risk factors associated with jigger infestation, design a self-adaptive machine-learning model for early detection, implement the model, and evaluate its effectiveness. Data were collected from 226 respondents, comprising 203 household/community members and 23 healthcare workers, and were used to identify risk factors and inform the model’s design. The image-based component used digital foot images representing two classes: Jigger and Non-Jigger. The dataset contained 20 images, including 14 Jigger and 6 Non-Jigger images. Images were cleaned, screened, standardized, and preprocessed before being used to train and evaluate a Convolutional Neural Network (CNN). The wider self-adaptive system combined the CNN prototype with an explainable adaptive risk component that incorporated validated case outcomes and risk parameters to generate LOW, MODERATE, and HIGH risk classifications. The model also incorporated feedback and a governed retraining pathway in which validated outcomes were used for model updating, while candidate models required approval before production deployment. The CNN prototype achieved 69.8% accuracy, 70.0% precision, 70.0% recall, and a 69.0% F1-score. A recorded adaptive-training attempt demonstrated the importance of data-quality safeguards because training could not proceed when both confirmed outcome classes were unavailable. The findings demonstrate initial technical feasibility of integrating image recognition, risk assessment, and adaptive monitoring within a single jigger-management platform, and longer-term field studies are needed to assess performance under changing community conditions.

Keywords: machine learning, Tunga penetrants, artificial intelligence, self-adaptive model, jigger infestation

DOI https://doi.org/10.33803/ma87eqPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ma87eq ↗

12.A Simulation-First Edge AI Framework for Agricultural Decision Support with a Quantization Error BudgetAbstract

John Nyamu Kibe^1*⁠iD, Morris Kimathi Mwangi¹⁠iD, Samuel Macharia Kang'iri^1*⁠iD

Low-cost weather nodes are cheap enough to deploy per field, but the logic that converts weather into irrigation and crop protection actions is normally a physically based model whose state variables such a node cannot measure. This paper frames embedded agricultural decision support as policy distillation: a physically based reference model supplies the decisions, and a classifier on an STM32 microcontroller reproduces them from what the node can sense. The reference couples a daily soil water balance, driven by FAO-56 reference evapotranspiration, to a thermal time and humid hour pest pressure model, forced by a stochastic weather generator fitted to central Kenyan climatology. Labels for irrigation timing, irrigation depth, fertilizer timing and pest risk are thresholds on simulated latent state, while the learner observes air temperature, humidity, wind speed, wind direction and barometric pressure. The task is therefore latent state reconstruction under partial observation, and each decision head's observability ceiling is reported separately from model error. A key contribution is the explicit separation of observability error: the irreducible disagreement caused by latent agricultural states unavailable to the sensor node from model error, representing the learner's failure to approximate the reference policy. Logistic regression is the first candidate because its decision at a fixed threshold depends only on sign of an affine score, so sigmoid approximation cannot alter hard decision and deployment error reduces to a margin condition: a sample changes class only if its margin falls below the score perturbation induced by Q15 coefficient quantization. Evaluation uses blocked temporal splits with an embargo, Matthews correlation and Brier score against majority class baseline, and Monte Carlo propagation of sensor tolerances. Irrigation timing reaches 98% accuracy in agreement with the reference policy, irrigation volume 85%, fertilizer timing 87%, and pest risk 86%, gap attributable to unobserved phenological state rather than model capacity.

Keywords: edge AI, fixed-point quantization, policy distillation, FAO-56 reference evapotranspiration, agricultural decision support

DOI https://doi.org/10.33803/y448fyPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/y448fy ↗

13.Assesment of Multilingual Automatic Speech Recognition for Swahili, Kikuyu and Kalenjin LanguagesAbstract

Joseph Muguro^1*⁠iD, Ciira Maina¹⁠iD, Faith Chege¹, Fancy Cherotich Koech¹

Language barriers continue to limit equitable access to digital and AI-assisted healthcare services, particularly for speakers of indigenous and low-resource languages. While Automatic Speech Recognition (ASR) has achieved significant progress for high-resource languages, its performance remains insufficiently explored for many Kenyan local languages due to limited and inconsistent speech datasets. This study investigates the development and evaluation of a multilingual ASR system for three languages spoken in Kenya: Swahili, Kikuyu (Gĩkũyũ), and Kalenjin. The study focuses on assessing the capability of a unified ASR model to recognize speech across these linguistically diverse languages while addressing the challenges associated with low-resource language processing. Speech datasets are acquired, prepared, and standardized for model fine-tuning and evaluation. The performance of the developed ASR system is assessed using Word Error Rate (WER) and Character Error Rate (CER), enabling comparative analysis of recognition accuracy across the three languages. The study further considers practical deployment requirements, including computational efficiency and suitability for real-world multilingual applications. By providing a focused evaluation of ASR performance for Swahili, Kikuyu, and Kalenjin, this research contributes to the growing body of work on speech technologies for underrepresented African languages. The findings are expected to identify language-specific recognition challenges and establish a foundation for integrating reliable multilingual speech interfaces into AI-assisted healthcare and other accessibility-focused applications.

Keywords: automatic speech recognition, low-resource languages, AI-Assisted healthcare

DOI https://doi.org/10.33803/tutkn5Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/tutkn5 ↗

14.Beyond Aggregate Indicators: Evaluating the Contextual Sensitivity of AI Readiness Frameworks for East African ContextsAbstract

Tevina Osewe^1*

Artificial intelligence (AI) is increasingly recognised as a tool for economic development, improving public services, and addressing complex social challenges across Africa making AI readiness an important consideration. AI readiness can be broadly defined as the capacity to effectively adopt, deploy and govern AI to realise its benefits. Existing assessments of AI readiness rely on frameworks that differ in conceptualisation and measurement of conditions required for effective AI adoption raising questions about whether existing frameworks adequately reflect contextual realities affecting AI readiness in East Africa. This study systematically reviews existing AI readiness frameworks to determine the extent to which contextual realities relevant to East Africa are accounted for. A structured document review was conducted using Google and official institutional sources with searches focused on “AI readiness framework” and “AI preparedness” for publications from 2019 to 2026. Four frameworks were selected based on institutional relevance and suitability for comparative analysis: Oxford Insights’ Government AI Readiness Index, UNESCO’s AI Readiness Assessment Methodology, Smart Africa’s AI Blueprint, and the ITU’s AI Readiness Framework. The frameworks are compared across five analytical dimensions: infrastructure, governance, institutional and implementation capacity, socioeconomic conditions, and human and technical capacity. Preliminary analysis indicates that existing frameworks recognise key structural and contextual dimensions relevant to AI readiness including infrastructure, governance, skills, economic conditions, and institutional capacity. However, they differ in the depth, granularity and methods through which these factors are assessed. Standardised indicators may not fully capture variations in infrastructure reliability and affordability, institutional implementation capacity, socioeconomic inequalities or the accessibility and relevance of human and technical expertise. UNESCO and ITU also demonstrate efforts to incorporate contextual and regional considerations. The study argues for greater attention to context-sensitive operationalisation in AI readiness assessment and contributes to developing more meaningful approaches to measuring AI readiness beyond aggregate indicators of technological capacity.

Keywords: artificial intelligence, AI readiness, Africa, responsible AI, AI governance, digital infrastructure, AI policy

DOI https://doi.org/10.33803/523xaePublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/523xae ↗

15.Beyond Prediction: Explaining Health Disparities in Diabetes Risk Through Fairness-Aware Machine Learning and Causal InferenceAbstract

Maurice Wanyonyi^1*⁠iD, Jacqueline Akelo Gogo¹⁠iD

Diabetes remains a major public health challenge, with substantial disparities in prevalence and outcomes across demographic and socioeconomic groups. Although machine learning has demonstrated promise for diabetes risk prediction, existing approaches have largely emphasized predictive accuracy, with limited attention to explainability, algorithmic fairness, and heterogeneous effects of modifiable risk factors. This study aimed to develop an integrated fairness-aware causal machine learning framework to improve understanding of diabetes risk disparities. Using the CDC BRFSS Diabetes Health Indicators dataset (n = 253,680), five machine learning algorithms, Logistic Regression, Random Forest, XGBoost, LightGBM, and CatBoost, were evaluated using accuracy, precision, recall, F1 score, ROC AUC, and Brier score with 95% bootstrap confidence intervals. SHAP was used for model explainability, fairness was assessed across sex, age, income, and education using demographic parity and equal opportunity metrics, and causal machine learning was applied to estimate heterogeneous treatment effects of modifiable risk factors. CatBoost achieved the highest overall predictive performance (accuracy = 0.8518, ROC AUC = 0.8250, Brier score = 0.1057), while Logistic Regression achieved the highest recall (0.7273). SHAP identified general health status, fruit consumption, BMI, age, and smoking as the most influential predictors. Fairness analysis revealed substantial disparities, including extremely low recall among adults aged 18–34 (0.015) and socioeconomic gradients in accuracy. Causal analysis demonstrated heterogeneous treatment effects, particularly for physical activity, with CATE increasing from 0.356 in the lowest-income group to 0.975 in the highest-income group. An integrated approach combining prediction, explainability, fairness assessment, and causal inference reveals disparities and heterogeneous intervention effects that conventional predictive modelling may overlook. These findings support equity-focused and more targeted diabetes prevention strategies and demonstrate the potential of responsible AI to address health disparities.

Keywords: algorithmic fairness, machine learning, explainable artificial intelligence, causal machine learning, diabetes prediction

DOI https://doi.org/10.33803/9efepjPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/9efepj ↗

16.Development of an Edge-AI Based Multimodal Framework (EM2-PNET) For Pneumonia Detection in Resource-Constrained SettingsAbstract

Francisco Mbugua Ngigi^1*⁠iD, Franklin Muriuki Manene^1*, Ciira Maina¹

Pneumonia remains a leading cause of mortality in resource-constrained settings, where limited access to trained clinicians and diagnostic imaging delays timely detection. This study presents EM2-PNET, an edge-AI based multimodal framework designed to enable low-cost, point-of-care pneumonia screening through synchronous acquisition of lung acoustics, body temperature, and photoplethysmography-derived oxygen saturation on a single embedded platform. A FreeRTOS-based, dual-core microcontroller architecture was designed and implemented to acquire the three modalities concurrently, incorporating adaptive sensor gain control, fault-tolerant data storage, and continuous system-health monitoring to support field reliability. Empirical validation demonstrated that inter-modality timestamp synchronization remained bounded within the system's logging interval, and that acoustic frequency response and temperature/oxygen-saturation sensing pipelines were functionally validated against reference instrumentation, with calibration refinement ongoing. A systematic, evidence-based investigation into heart-rate estimation identified signal-quality limitations as the governing constraint, distinct from algorithmic design, and is reported as a characterized limitation rather than an unresolved defect. In parallel, exploratory analysis of two public clinical datasets, the ICBHI 2017 Respiratory Sound Database and MIMIC-IV Emergency Department vitals, was conducted to inform development of a hybrid convolutional neural network and multilayer perceptron architecture intended to fuse acoustic and physiological features for pneumonia classification. This analysis revealed a patient-level class imbalance and demographic confound within the acoustic dataset's pneumonia category, informing a revised, literature-grounded patient-level cross-validation strategy, while diagnosis-linked vitals from the clinical dataset exhibited abnormal-value prevalence consistent with expected clinical presentation. The study was conducted under formal institutional ethics approval. Collectively, this work establishes a validated, low-cost hardware acquisition platform and a rigorously characterized data foundation for subsequent model development, representing a methodologically transparent contribution toward deployable multimodal diagnostic support for pneumonia detection in underserved healthcare settings.

Keywords: edge artificial intelligence, multimodal sensing, pneumonia diagnostics, point-of-care, resource-limited settings, model optimization, wearable health, low-power inference

DOI https://doi.org/10.33803/k62v32Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/k62v32 ↗

17.Hybrid Edge-Cloud AI for Livestock Behaviour and Health Monitoring Using an AIoT Wearable CollarPoster

Patty Mailu Mutinda^1*⁠iD, David M Ngari¹, Samwel Kamau¹, Benson Nganga¹, Jeff Kamau¹, Leone Njiru¹, Brian Koech¹, Samuel Macharia Kang'iri^1*⁠iD

Background. Smallholder dairy farmers can miss changes in cow behaviour and breeding cues when relying periodic observation. Missed estrus may delay breeding and affect productivity. A low-cost collar that tracks behaviour and temperature trends could support earlier follow-up and build longitudinal livestock records relevant to future credit access.

Methods. An AIoT collar integrating a 9-axis IMU and a DS18B20 surface temperature sensor (on shaved neck skin) executes local behavior classification on an STM32F411 microcontroller. Summaries and oestrus alerts are transmitted via LoRa to a base station gateway, which forwards data to a cloud backed via an A7670E 4G LTE Cat 1 module using MQTT. A pilot at DeKUT School Farm cows will benchmark edge inferences against farm attendant observer ethograms.

Results. The system remains in development and has not yet been field-tested. The approximately 90% classification accuracy and 21-day battery life are design targets. No measured animal outcomes or credit assessments are available at the moment.

Conclusions. The collar is intended to help farm attendants notice behavioural trends and follow up on possible breeding or health concerns. If validated, longitudinal livestock profiles could support small scale farmers' credit applications; the system will not provide a confirmed diagnosis

Keywords: livestock asset valuation, precision livestock farming, TinyML, AIoT

DOI https://doi.org/10.33803/j5puehPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/j5pueh ↗

18.Implementing an Ensemble Model Based on Shifted Window Transformers for Multi-Class Lumpy Skin Disease DetectionFull paper

George Mwangi Muhindi^1*⁠iD, Geoffrey Mariga Wambugu¹, Aaron Mogeni Oirere²

Lumpy Skin Disease (LSD) is a significant infectious cattle disease that negatively impacts livestock productivity, animal health and trade. While the clinical examination, PCR and serological testing are effective methods, they can be expensive, require sophisticated equipment and take a long time to get results and are therefore limited in rapid screening in resource-limited environments. The overall objective of this study was to design, deploy, and test an explainable multi-level deep learning model using Shifted Window Transformers for automated grading and identification of cattle affected with LSD. The model comprised three components: the use of Vision Transformer (ViT), ConvMixer and Swin Transformer to learn global semantic relationships, local lesion textures, and hierarchical spatial features, respectively, and the inclusion of Gradient-weighted Class Activation Mapping (Grad-CAM) for visualizing model predictions. A total of 4,190 cattle images categorized as Normal, Mild, and Severe LSD were used, with the dataset partitioned using an 80:10:10 ratio into 3,352 training, 419 validation, and 419 unseen independent test images. The proposed three-tier ensemble achieved an overall test accuracy of 97.37%, with weighted precision of 97.38%, recall of 97.37%, and F1-score of 97.37%, correctly classifying 408 of the 419 test images. The standalone ViT-B/16, ConvMixer, and Swin Transformer Tiny models achieved accuracies of 93.88%, 56.56%, and 93.32%, respectively. Veterinary field validation yielded 88.33% agreement of experts with the model, and a Cohen's kappa coefficient of 0.823, while the Grad-CAM explanations had a mean expert alignment score of 4.02/5 with 77.33% having high or excellent expert alignment. Future work should focus on expanding the dataset with more diverse field images. There is a need to explore approaches to optimizing the developed model for deployment on resource limited devices like tablets, edge computing devices, and smartphones.

Keywords: artificial intelligence, deep learning, shifted window transformer, explainable AI, lumpy skin disease, veterinary diagnostics

DOI https://doi.org/10.33803/fpt46ePublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/fpt46e ↗

19.Knowledge Distillation and Structured Pruning at Matched Footprint for Sub-50 KB Cattle Behaviour ClassificationFull paper

Bright Mutuma Gitari¹, Ying-Hsuan Chiang²⁠iD, Consolata Wanjiru Kiiru^3*⁠iD, Samuel Macharia Kang'iri^1*⁠iD

On-collar behaviour classification underpins automated oestrus and health monitoring in cattle, but accurate models exceed the memory a low-cost collar can afford. Knowledge distillation and structured pruning are both proposed as the remedy, but on-collar studies test them separately and without the control that makes both interpretable: a same-size network trained from hard labels. Using two public neck-accelerometer datasets of cattle, 6 Japanese Black beef cows (the beef dataset) and 30 pre-weaned calves (the calf dataset), a convolutional teacher is trained under leave-one-animal-out validation and compressed along three routes at a matched int8 footprint, with the frontier re-derived on the calf dataset. First, the released beef dataset contains cross-animal duplication: all 15 animal pairs share byte-identical blocks of varying signal, and leaving it in place inflates macro F1 by +0.076 even when each animal is scored only on the classes it shows (p = 0.0007). Second, the capacity deficit is almost nil: a 1.72 KB network trained from hard labels reaches 0.681 macro F1 against the 127.1 KB teacher’s 0.692, which caps what compression can be worth. Within that ceiling, neither route is reliably better than same-size training. Structured pruning gains +0.0130 at one seed (p = 0.0096, n = 42) but −0.0209 to +0.0145 over five seeds, collapsing for some animals at the smallest widths at three seeds. Fine-tuning it on the same schedule as the other routes removes most collapses and the margin with them (−0.0080 to +0.0038 over three seeds, never significant). Distillation gains are +0.0021 to +0.0086, significant at one seed. Third, on calves, where shrinking costs up to +0.077, pruning falls below same-size training at five of five seeds, significantly at two, so no ordering here is a property of the methods.

Keywords: knowledge distillation, structured pruning, TinyML, precision livestock farming, dataset integrity

DOI https://doi.org/10.33803/kg23r7Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/kg23r7 ↗

20.Offline Cervical Cytology Classification: A Comparative Evaluation of Deep Learning Architectures for Edge DeploymentPoster

John William Muga^1*⁠iD, Yusupha Ceesay², Edna Wairimu Mugo³, Thomas Makundi Kitavi⁴, Khanyisile Tapiwa Magagula⁵

Background. The severe shortage of cytopathologists and centralized laboratory infrastructure is a major challenge for cervical cancer in Sub-Saharan Africa as a leading cause of preventable deaths. Current Deep Learning models are based on cloud deployment, which causes a lot of latency and connectivity issues in rural clinics. In the present work, edge-focused deep learning architectures were optimized for the purpose of fully automatic, 100% offline cervical cell classification on conventional smartphones.

Methods. We trained and tested three deep learning architectures separately: EfficientNet-B0 (lightweight CNN), BiomedCLIP (vision-language model), and DINOv2 (self-supervised vision transformer). We evaluated them on the SIPaKMeD benchmark dataset, which consists of 4,049 annotated single-cell images from five clinical categories cropped from 966 Pap smear slides and annotated by experts. The model's performance was assessed using the accuracy and macro-averaged F1-scores of the 70:15:15 train/validation/hold-out split.

Results. The held-out test accuracy and test F1 score of EfficientNet-B0 are 94.71% and 0.9483, respectively. BiomedCLIP outperformed the other models, achieving 94.71% test accuracy (F1: 0.9480) and 98.10% training accuracy (F1: 0.9815). DINOv2's test accuracy was 94.03% and the test F1-score was 0.9411.

Conclusions. EfficientNet-B0 seems to be a more suitable choice for an architecture in the context of an offline, mobile-based cervical cancer cell classification platform. It can match the predictive power of larger foundation models while using a negligible fraction of the computational resources, and it can serve as a practical triage-assistive diagnostic tool for healthcare workers at primary care clinics without access to cloud resources or specialists.

Keywords: cervical cancer, point-of-care, deep learning, low-resource, SIPaKMeD, EfficientNet-B0, BiomedCLIP, DINOv2

DOI https://doi.org/10.33803/rjrpfrPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/rjrpfr ↗

21.Predicting Closing Stock Returns Using Sequential Artificial Neural NetworkFull paper

James Asumbo Otieno^1*, Dennis Cheruiyot²⁠iD, Simon Maina Mundia²⁠iD

Financial series, particularly daily equity prices and asset returns, are inherently characterized by non-stationarity, extreme volatility clustering, non-linear dynamics and complex chronological dependencies. Accurately modeling and predicting these financial series is crucial for active risk management, capital allocation and algorithmic trading. Traditional econometric models such as Autoregressive, Autoregressive Moving Average, Autoregressive Integrated Moving Average and static artificial neural networks frequently fail to capture the complex, non-linear and time-dependent structures inherent in financial time-series data due to their underlying assumptions . This study evaluates the out-of-sample forecasting performance of a Dynamic Long Short-Term Memory network against Static LSTM, Feed-Forward Artificial Neural Networks and traditional baselines and a Zero-Return benchmark across the "Magnificent Seven" technology equities from January 1, 2006, to January 1, 2026. Across the unseen 20% test partition, the Dynamic LSTM achieves superior performance, attaining an average root mean square error of 0.0187 and a Mean Absolute Error of 0.0182 on daily log-returns, a directional accuracy of 58.4%, and a Theil’s U statistic of 0.612, outperforming the zero-return baseline threshold (U= 0.707). Jarque-Bera tests on out-of-sample forecast residuals confirm non-normal, leptokurtic error distributions, driven by tail events and volatility clustering. The statistical superiority of DLSTM over classical benchmarks is confirmed via the Diebold-Mariano test . These findings demonstrate that recursive sliding-window state updates provide a robust quantitative decision-support framework for risk assessment and algorithmic execution strategies. The study recommends the integration of optimized dynamic sequential models into trading systems and portfolio risk management frameworks. The model should be used as a decision-support tool not as a guarantee of investment returns and suggests future research exploring alternative sequential mechanisms such as Gated Recurrent Units. Keywords: stock return prediction, Sequential artificial neural network, dynamic Long short-term Memory, deep learning, financial time series, closing stock returns, directional accuracy, algorithmic forecasting

Keywords: stock return prediction, sequential artificial neural network, dynamic long short-term memory, deep learning, financial time series, closing stock returns, directional accuracy, algorithmic forecasting

DOI https://doi.org/10.33803/9sc487Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/9sc487 ↗

22.Speed-Induced Dilation and Prototype Alignment in Federated Bearing Fault DiagnosisAbstract

Joseph Munene Disamalu^1*⁠iD, Morris Kimathi Mwangi¹⁠iD, Samuel Macharia Kang'iri^1*⁠iD

Speed variation across industrial edge nodes is usually treated as unstructured non-IID drift in federated fault diagnosis. Bearing characteristic frequencies are proportional to shaft speed, so a speed ratio ρ acts on the vibration signal as a time-axis dilation, placing client distributions on a one-parameter orbit. A 1D CNN is equivariant to translation but not to dilation, which accounts for the feature drift reported in cross-speed studies. Mapping the normalized squared envelope spectrum onto a logarithmic frequency axis converts the dilation into a translation of log ρ, restoring approximate equivariance without a tachometer. We then state a margin condition on the projection sphere: a nearest-prototype decision at an unseen speed survives when the induced displacement plus the class radius stays below half the minimum inter-prototype separation. FedProc enforces this during training. Each client runs an encoder, a classifier head, and a projection head. A contrastive prototype loss pulls each L2 normalized projection toward its class prototype and away from the others; the server refreshes prototypes each round by weighted averaging of client estimates. No discriminator, generator, or target-domain sample is needed, and the exchanged payload is C·k floats per client per round. A stochastic bearing model (impulse train convolved with a single degree of freedom resonance, random slip, load modulation, speed-dependent noise) generates paired federations at arbitrary ρ, so displacement, class radius, and prototype separation are measured directly rather than bounded. Monte Carlo sweeps place the accuracy knee at the measured margin violation. On JNU (600 to 1000 rpm), FedProc improves held-out accuracy over FedAvg, FedProx, and a federated adversarial baseline under leave-one-speed-out. On CWRU, where speeds differ by under 4 percent, the margin holds and the methods perform comparably, as predicted.

Keywords: federated learning, bearing fault diagnosis, speed-induced dilation, prototype contrastive alignment, simulation-based validation

DOI https://doi.org/10.33803/8judyePublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/8judye ↗

23.ST-RAG: Spatio-Temporal Retrieval-Augmented Generation for Behaviour-Grounded Virtual Agents in Immersive Clinical TrainingAbstract

Edwin Ngari^1*⁠iD, Samuel Macharia Kang'iri^2*⁠iD, Jean B. Byiringiro¹

Retrieval-augmented generation grounds language models in retrieved text, but text is a poor description of what a trainee is doing inside an immersive simulation. Current systems close this gap with hand-authored trigger rules that scale poorly and break when learners act unexpectedly. We present ST-RAG, an architecture that conditions both retrieval and generation on a rolling window of the learner's embodied behaviour. A Unity client streams speech together with gaze target and dwell time, interpersonal distance in proxemic bands, hand-object contact, and locomotion to a local server over WebSocket. Whisper transcribes the utterance while the behavioural window is discretised, by a fixed hand-designed scheme with scenario-independent dimensions and scenario-specific targets, into situational states and state transitions, which expand and filter a query over a local vector store of scenario, protocol, and patient-history documents. Because prefill cost grows linearly with library size and is incurred on every turn, prompt injection admits only a bounded scenario library at any latency target, whereas retrieval does not. A locally hosted instruction-tuned model returns a structured turn: the spoken response plus non-verbal control tags (affect, gaze aversion, guarding) that drive avatar blendshapes, closing the loop from learner behaviour to agent behaviour with no cloud dependency and no patient data leaving the machine. We instantiate ST-RAG in a wound-assessment scenario and compare it with text-only RAG, a gaze-only variant, and a scripted rule baseline on situational appropriateness, rated against a rubric by two blinded clinicians; retrieval preci- sion over a behaviour-annotated benchmark whose queries match in wording and differ only in behaviour; co-presence, by a published social-presence instrument; and end-to-end latency from end of learner speech to onset of agent speech, on a single consumer GPU.

Keywords: spatial-temporal RAG, artificial intelligence, embodied virtual agents, multimodal context fusion, VR clinical simulation

DOI https://doi.org/10.33803/y9qsvbPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/y9qsvb ↗

24.Accounts Payable Management Practice and Financial Performance of Public TVET Institutions in Kenya: Evidence from Baringo CountyAbstract

Rebecca Jepkosgei Chelanga^1*

Public Technical, Vocational Education and Training (TVET) institutions in Kenya finance a substantial share of instructional materials, equipment, and workshop consumables through supplier credit, making accounts payable central to institutional financial control. This study assessed the effect of accounts payable management practice on the financial performance of public TVET institutions in Baringo County. Data were collected from 118 finance, procurement, and administrative staff from the 16 public TVET institutions in the county (81.9% response rate), with accounts payable management practice operationalized across four dimensions: invoice processing, payment scheduling, adherence to payment terms and digitalization of accounts payable processes. The construct showed good internal consistency (α = 0.879) and was highly rated by participants (M = 4.49, SD = 0.63). The Pearson correlation coefficient showed a relatively strong and significant relationship between accounts payable management practices and financial performance (r = .536, p < .001). A simple linear regression analysis revealed that accounts payable management practices predicted financial performance, R² = .287, adjusted R² = .281, F(1, 116) = 46.76, p < .001, and was the strongest and only significant working capital predictor in a supplemental model after controlling for cash management, inventory management and accounts receivable management as well as intervening institutional variables (β = .482, p < .001). Digitalization of accounts payable processes was rated as a consensus factor (M = 4.72, SD = 0.451) These results indicate that accounts payable management is not a peripheral administrative function in public TVET institutions but a demonstrable driver of institutional financial performance. The paper recommends that institutions formalize invoice verification and payment scheduling procedures, deepen digitalization of accounts payable processing, and that sector regulators embed accounts payable performance standards into financial management guidance for public training institutions.

Keywords: accounts payable management, invoice processing, payment scheduling, trade credit, digitalization, financial performance

DOI https://doi.org/10.33803/q9cbkcPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/q9cbkc ↗

25.Affordable Housing Driving Sustainable Tourism Through Homestays and Short-Term Rentals in KenyaAbstract

Peter Ogalo Ogweyo^1*⁠iD, Peninah Wanjiku Chege¹

This study examines the potential of affordable housing to drive sustainable tourism development in Kenya, focusing on homestays, short-term rentals, and emerging accommodation models. The central premise is that affordable housing transcends its role as a welfare concern; it can function as a core component of tourism infrastructure and a strategic tool for shaping destination growth. To guide the analysis, the study develops the Affordable Housing–Tourism Development Nexus (AHTDN) Model, a conceptual framework illustrating the dynamic linkages between residential homes and visitor lodging. Employing a policy and concept-based methodology through document review, the research analyzes existing laws, including Kenya's Tourism Act (2011) and Affordable Housing Act (2024), alongside international policy lessons. The findings reveal that affordable housing can enhance tourism by increasing accommodation capacity, empowering local communities through homestay ownership, supporting rural tourism, and providing stable housing for tourism workers. However, the study also identifies significant risks, including the displacement of residents and increased rent prices when short-term rentals proliferate without regulation. To mitigate these risks and harness the synergies, the study proposes an integrated policy framework for Kenya. This framework recommends risk-based regulation for short-term rentals, the use of digital registries for data-driven management, and the inclusion of housing in destination planning. It emphasizes land-use planning that balances tourism needs with the preservation of residential housing stock. In conclusion, the study urges Kenya to adopt a joined-up approach that links housing and tourism policy, ensuring that housing fulfills its core social purpose while responsibly supporting local economic development and sustainable tourism.

Keywords: affordable housing, tourism development, homestays, short-term rentals, community tourism

DOI https://doi.org/10.33803/ywwh53Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ywwh53 ↗

26.Does Climate Change Vulnerability Shape Structural Transformation in Africa? A Dynamic Panel Data TestAbstract

Jerry Ogutu Sumba^1*⁠iD

Climate vulnerability continues to increase while structural transformation remains low in Africa, estimated at around 10% of GDP, which is below levels achieved by other developing regions such as Asia and Latin America. This low industrialization has been stated to be compounded by climate change vulnerability, which disrupts agricultural production, infrastructural development, investment, and industrial activities. This study examined how climate vulnerability shapes structural transformation in Africa, focusing on whether vulnerability-transformation exhibits a non-linear threshold relationship, an area previous studies have not explored. The Cross-Sectionally Augmented Autoregressive Distributed Lag (CS-ARDL) and Dynamic Panel Threshold (kink) models were employed on panel data for 45 African countries spanning 2000-2024 for analysis. The structural transformation was measured by the net manufacturing output as a percentage of GDP, while climate vulnerability was captured as the inverted ND-GAIN vulnerability sub-index (0-100). The study found that rising climate vulnerability significantly reduced structural transformation, with a unit increase in climate vulnerability associated with an 11.8 percentage-point decline in manufacturing output. Similarly, 58.4 ND-GAIN was estimated as the threshold level beyond which every additional unit increase in vulnerability reduces manufacturing output by a further 6.1 percentage points. The study further found that approximately 62% of the sampled countries surpassed the estimated threshold, implying widespread exposure to a high-damage regime. These findings highlight the need for integrating climate adaptation into industrial development strategies in the region. The study recommends channeling climate finance toward the development of climate-resilient manufacturing infrastructure, production systems, and technologies to strengthen early-warning and adaptation systems that can protect industrial supply chains. Second, the study recommends the use of 58.4 as a climate vulnerability benchmark for monitoring and intervention. Green industrialization should also be promoted through renewable energy and climate-smart technology investment, while encouraging economic diversification to reduce dependence on climate-sensitive sectors and sustain industrial upgrading.

Keywords: structural transformation, climate change vulnaribility, manufacturing value added, ND-GAIN index, dynamic panel threshold, panel threshold kink model, industrialization

DOI https://doi.org/10.33803/bne8u5Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/bne8u5 ↗

27.Effect of Cross-cultural Communication on the Performance of Multinational Corporations in Kericho County, KenyaAbstract

Judith C Mutai^1*

The performance of multinational corporations (MNCs) in culturally diverse workplaces depends in part on how managers and employees communicate across language and cultural boundaries. This study examined the effect of cross-cultural communication on the performance of MNCs in Kericho County. It was grounded in Social Exchange Theory and adopted a quantitative cross-sectional survey design. Structured questionnaires were administered to managers and subordinates in tea-sector MNCs through stratified simple random sampling. Of the 426 questionnaires distributed, 282 were completed, giving a 66.2% response rate. Cross-cultural communication was measured through collaborative communication, communication adaptation, language inclusivity, conflict resolution, and cultural awareness programs. Performance was measured through cultural adaptability, strategic goal achievement, service or product quality, operational efficiency, and employee productivity. Spearman’s rank correlation analysis demonstrated a moderate positive association between cross-cultural communication and perceived performance (ρ = .500, p < .001). The one-predictor OLS model with HC3 robust standard errors was significant (Wald F(1, 280) = 106.80, p < .001), accounting for 45.7% of the variability in performance ratings. Bivariate analyses revealed communication was significantly positively associated with performance (B = 0.645, β = .676, p < .001, 95% CI [0.522, 0.768]). Collaborative communication (M = 4.66) achieved the highest item ranking, whereas language inclusivity was the lowest-rated factor with the broadest range of responses (M = 4.47, SD = 0.67). Given the cross-sectional nature of the data and reliance on self-reports, the results demonstrate associations and mainly reflect the views of subordinates. Managers should preserve collaborative communication while strengthening language inclusivity through multilingual instructions and accessible grievance procedures.

Keywords: cross-cultural communication, multinational corporations, organizational performance, Kericho county, tea sector

DOI https://doi.org/10.33803/n2fbfePublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/n2fbfe ↗

28.Effect of Financial Literacy on Financial Risk Management Among Small and Medium Enterprises in Nakuru County, KenyaAbstract

Lynn Jepkemei Yatich^1*

Small and Medium-sized Enterprises (SMEs) are critical drivers of economic growth, employment creation, and innovation in Kenya. However, a significant proportion of these enterprises fail within their first five years of operation due to inadequate financial management capacity and limited ability to handle financial risks. This study examined the effect of financial literacy on financial risk management among SMEs in Nakuru County, Kenya. Specifically, the study evaluated the influence of financial awareness, financial skills, and financial attitudes on financial risk management practices. Grounded in Financial Capability Theory, Human Capital Theory, and Mental Accounting Theory, the study adopted a survey research design utilizing a predictive correlational approach. The target population comprised 336,661 registered SME owners and managers across eleven sub-counties in Nakuru County. A sample size of 100 respondents was selected using Nassiuma’s formula through proportionate stratified random sampling. Primary data were gathered using structured self-administered questionnaires and analyzed using descriptive statistics, Pearson correlation, and multiple linear regression. The findings revealed that financial literacy significantly and positively impacts financial risk management among SMEs (R^2 = 0.709, F = 65.82, p < 0.05). Financial skills demonstrated the strongest influence (β= 0.485, p < 0.05), followed by financial awareness (β= 0.325, p < 0.05) and financial attitudes (β = 0.211, p < 0.05). The study concludes that technical financial skills, coupled with high financial awareness and balanced risk attitudes, enhance the capacity of SMEs to mitigate financial shocks. The study recommends that public institutions, financial service providers, and SME support organization implement targeted financial capability programs to boost business resilience and sustainability.

Keywords: financial literacy, financial awareness, financial skills, financial attitudes, financial risk management, small and medium enterprises

DOI https://doi.org/10.33803/gwavvrPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/gwavvr ↗

29.Effect of Mentorship and Coaching on Organisational Performance in Faith-based Non-governmental Organisations in Nairobi CountyAbstract

Sharon Chepkemoi^1*

Leadership continuity and human resource management are critical for organisational effectiveness, yet Faith-Based Non-Governmental Organisations (FBNGOs) in Kenya face persistent challenges including staff turnover, limited investment in employee development, and unplanned leadership transitions. While mentorship and coaching have been extensively studied in corporate settings, limited empirical evidence explains how these practices operate within Kenya's resource-constrained, values-oriented faith-based context. This study therefore investigates the effect of mentorship and coaching on organisational performance in FBNGOs in Nairobi County, applying Relay Succession Planning Theory to explain these practices as mechanisms for leadership continuity and knowledge transfer. A quantitative cross-sectional survey design was employed, targeting 73 FBNGOs selected using the Yamane formula from a population of 90 registered organisations. A purposive sample of 219 respondents comprising (junior management, administrative/HR, and senior management staff) completed pre-tested structured questionnaires. Data were analysed using multiple regression and descriptive statistics via SPSS. Because the study instrument operationalised mentorship and coaching as a combined predictor, results are reported for the joint construct. The combined mentorship and coaching construct demonstrated a positive, statistically significant relationship with organisational performance (r = 0.642, p < 0.001; β = 0.337, t = 2.526, p = 0.015). The study concludes that mentorship and coaching, considered jointly, significantly enhance organisational performance in FBNGOs. The findings refine Relay Succession Planning Theory by demonstrating that these practices serve as practical relay mechanisms for transferring institutional knowledge, leadership capabilities, and organisational values to emerging leaders. The study recommends that FBNGOs formalise mentorship programmes through structured mentor assignments, mentors-training, and regular feedback mechanisms. Additionally, organisations should strengthen resource mobilisation and donor engagement to sustain mentorship and coaching beyond short-term project funding while deliberately developing leadership pipelines for future succession. These findings have broader implications for non-profit and faith-based organisations across Kenya facing similar leadership continuity and human-resource challenges.

Keywords: mentorship and coaching, organisational performance, faith-based non-governmental organisations

DOI https://doi.org/10.33803/pqqef7Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/pqqef7 ↗

30.From Policy Silos to Policy Coherence: Integrating Tourism and Urban Development Policies in Nairobi and Mombasa City CountiesAbstract

Rodgers Isaac Kisiang'ani^1*⁠iD, Márk Miskolczi²⁠iD

Urban tourism is increasingly recognized as a strategic component of urban development, yet its governance in many cities remains fragmented across sectoral boundaries, leading to policy inconsistencies and coordination challenges. Integrating tourism policy into urban development policy has emerged as an important approach to enhancing cross-sectoral policy coherence and promoting coordinated urban governance. However, empirical evidence on how this integration is achieved in African cities and the extent to which it fosters coherent policymaking across sectors remains limited. This study examines how Nairobi and Mombasa City County governments integrate tourism policy into urban development and the impact on cross-sectoral policy coherence. This analysis of 45 policy documents from 2010 to 2026, using PRISMA 2020 and Qualitative Comparative Analysis, assesses policy alignment, institutional arrangements, planning, and implementation. Results show varying levels of integration; stronger ties improve policy alignment and coordination, while weaker ties lead to fragmentation. The study offers a framework for evaluating tourism-urban policy integration and insights for policymakers, planners, and tourism authorities to promote coordinated urban governance.

Keywords: urban tourism, tourism policy integration, urban development policy, cross-sectoral policy coherence

DOI https://doi.org/10.33803/7g9msgPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/7g9msg ↗

31.Influence of Budgeting Practices on Financial Sustainability of County Governments in KenyaAbstract

Harrison Bwire Junge^1*⁠iD, Stephene Oloo Magadi²⁠iD, Symon Kiprop²

Financial sustainability remains a concern for county governments in Kenya because it determines their capacity to mobilize and utilize resources, meet financial obligations, sustain service delivery, and implement programmes. Although previous studies have examined budgeting and fiscal performance in county governments, limited attention has been given to the relationship between budgeting practices, operationalized as budget formulation, revenue forecasting, expenditure control, and budget monitoring, and financial sustainability. This study examined the relationship between budgeting practices and financial sustainability of county governments in Kenya, addressing the gap. Guided by fiscal federalism theory, the study used a descriptive, correlational design and a quantitative approach. The study operationalized financial sustainability as the capacity to mobilize and use resources, meet financial obligations, sustain service delivery, and support development programs. We collected data through questionnaires administered to county officials. Of 341 questionnaires, 280 were returned, for an 82.1% response rate. The instrument demonstrated content validity (S-CVI = 0.91) and internal consistency, with Cronbach’s alpha coefficients of 0.847 for budgeting practices and 0.879 for financial sustainability; overall reliability was 0.878. We analyzed the data using descriptive statistics, Pearson correlation, and simple linear regression. Findings showed a strong, positive, and statistically significant relationship between budgeting practices and financial sustainability (r = 0.784, p < 0.01). Regression results indicated that budgeting practices explained 61.5% of the variation in financial sustainability (R² = 0.615). The regression coefficient was positive and statistically significant (β = 0.784, t = 21.06, p < 0.001), leading to rejection of the null hypothesis. Findings support fiscal federalism theory in explaining budgeting and financial sustainability within Kenya’s devolved system. The study recommends strengthening budget formulation, revenue forecasting, expenditure controls, and budget monitoring, alongside capacity building for finance personnel. The study contributes evidence linking multidimensional budgeting practices to financial sustainability in county governments.

Keywords: budgeting practices, financial sustainability, county governments, budget implementation, expenditure control

DOI https://doi.org/10.33803/baswpgPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/baswpg ↗

32.Influence of Change Management Practices on Organizational Performance at the Kenya Revenue AuthorityAbstract

Jackson Kimeu Kyalo^1*

Organizational reforms in public sector institutions do not automatically translate into improved organizational performance, particularly where employees are insufficiently prepared for change, communication is unclear, participation is limited, and mechanisms for evaluating change outcomes are weak. This study investigates the influence of change management practices on organizational performance at the Kenya Revenue Authority (KRA). Specifically, the study examines the influence of change readiness, change communication, change engagement, and change evaluation on organizational performance. The study is anchored on Lewin’s Three-Step Change Model, Kotter’s Eight-Step Change Model, and the Prosci ADKAR Model, which collectively explain how organizations prepare employees for change, communicate reform initiatives, facilitate participation, institutionalize new practices, and sustain performance improvement. Organizational performance is examined through revenue collection efficiency, service delivery, employee productivity, compliance levels, and policy implementation effectiveness. The study adopts a mixed-methods research design. The target population comprises 4,000 employees at KRA Headquarters, from which a sample of 364 respondents is determined using Yamane’s formula at a 5% margin of error. Stratified random sampling will be used to select quantitative survey respondents, while purposive sampling will identify 20–25 participants for qualitative interviews. Data will be collected using structured Likert-scale questionnaires and semi-structured interviews. Quantitative data will be analyzed using descriptive statistics, Pearson correlation, and multiple regression analysis, while qualitative data will be analyzed thematically. The study is expected to contribute empirical and contextual insight into how structured change management practices can support organizational performance in public sector revenue administration, while providing practical guidance for managing organizational reforms within complex public institutions.

Keywords: change management practices, organizational performance, change readiness, change communication, change engagement, change evaluation, Kenya revenue authority

DOI https://doi.org/10.33803/banrgrPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/banrgr ↗

33.Influence of Health Programmes on Employee Performance Among Fairtrade-certified Flower Farms in Nakuru County, KenyaAbstract

Julia Jerono Koech^1*

Employee health is an important component of employee welfare and can influence employees' capacity to maintain productivity, quality, attendance and commitment. This study examined the influence of health programmes on employee performance among Fairtrade-certified flower farms in Nakuru County, Kenya. A descriptive-explanatory research design was adopted, and quantitative data were collected from employees using structured questionnaires. Data from 199 usable responses were analysed using descriptive statistics, Pearson correlation and multiple regression analysis. Health programmes recorded the highest composite mean among the welfare constructs, with a mean of 3.643 (SD = 0.679). Pearson correlation analysis established a strong, positive and statistically significant relationship between health programmes and employee performance (r = .620, p < .001). Multiple regression analysis established that health programmes made a positive and statistically significant contribution to employee performance (B = .335, β = .344, t = 4.648, p < .001). The overall regression model was statistically significant, F(4, 194) = 41.985, p < .001, explaining 46.4% of the variance in employee performance. The null hypothesis that health programmes have no statistically significant influence on employee performance was rejected. Health programmes emerged as the strongest individual predictor of employee performance. The study concludes that health programmes are a critical component of employee welfare and should be strengthened within Fairtrade-certified flower farms.

Keywords: health programmes, employee performance, employee welfare, occupational health, fairtrade, Nakuru county, Kenya

DOI https://doi.org/10.33803/epxub5Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/epxub5 ↗

34.Influence of Prescriptive Strategic Mode on Competitive Advantage of Retail SMEs in Homa Bay Town, KenyaAbstract

Odera Awuor Dorothy^1*, Daisy Bowen¹, Mongáre Omare¹

Retail SMEs in Homa Bay Municipality, Kenya, face significant challenges related to business survival and growth. Limited research has examined how prescriptive strategic practices influence competitiveness in county-level retail contexts. Therefore, the study examined the influence of prescriptive strategic mode on competitive advantage of retail SMEs in Homa Bay Town, Kenya. An explanatory cross-sectional design targeted 105 retail SME owners/managers through a census. Data were collected using a structured questionnaire validated through a pilot study, with a Cronbach’s alpha coefficient of 0.765. Descriptive statistics, including mean and standard deviation, and simple linear regression analysis were used. The results revealed that the regression model was statistically significant (F = 150.279, p < .001). Prescriptive strategic mode had a positive and statistically significant association with competitive advantage (β = .713, p < .05), explaining 59.3% of the variance in competitive advantage (R² = .593). All four dimensions formalization of planning (M = 3.89), stakeholder engagement (M = 3.90), organizational learning (M = 4.06), and environmental scanning (M = 4.15) were highly practiced, with environmental scanning recording the highest mean. The study concludes that prescriptive strategic mode is positively associated with competitive advantage, with structured planning and environmental scanning being important strategic practices among retail SMEs. Retail SMEs should formalize planning routines, institutionalize environmental scanning, and strengthen stakeholder engagement. Local agencies should facilitate strategic planning training, while financial institutions should support structured business planning. Keywords: Prescriptive Strategic Mode, Competitive Advantage, Retail SMEs, Strategic Planning, Environmental Scanning Keywords: Prescriptive Strategic Mode, Competitive Advantage, Retail SMEs, Strategic Planning, Environmental Scanning

Keywords: prescriptive strategic mode, competitive advantage, retail SMEs, strategic planning, environmental scanning

DOI https://doi.org/10.33803/zap637Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/zap637 ↗

35.Performance Appraisal Practices, Industry Regulation and Performance of Insurance Companies in Nairobi City, KenyaAbstract

Emmy Jepchumba Kisa^1*, Ronald Chepkilot², Emily Cheposero Tumwet²

Performance appraisal remains a critical determinant of organizational performance in service-intensive and highly regulated sectors like insurance, but the strength of this relationship may depend on the regulatory environment within which appraisal systems operate. This study examined the relationship between performance appraisal practices and the performance of insurance companies in Nairobi City, Kenya with industry regulation being the moderating variable. Adopting a descriptive cross-sectional research design, the study targeted 116 Human Resource Officers and Branch/Operations Managers from 58 registered insurance companies operating in Nairobi City. Data were collected through structured questionnaires and analyzed using descriptive statistics, Pearson correlation, simple and moderated linear regression. The findings revealed that performance appraisal practices were perceived as moderately implemented but developmentally weak, with mean scores ranging from (Mean 3.45 to 3.92) on a five-point scale, while industry regulation recorded the highest mean of all constructs (Mean = 3.90), reflecting a strong and consistent regulatory influence on human resource functions. Correlation analysis established significant positive relationships between performance appraisal and organizational performance (r = 0.553, p < 0.01), industry regulation and organizational performance (r = 0.568, p < 0.01), and performance appraisal and industry regulation (r = 0.476, p < 0.01). Regression analysis further confirmed that performance appraisal had a statistically significant positive influence on organizational performance, and that the interaction between performance appraisal and industry regulation was statistically significant, meaning industry regulation moderates the relationship between performance appraisal and organizational performance. The study concludes that performance appraisal practices are positively linked to organizational performance, and that this relationship is strengthened when appraisal practices are aligned with industry regulatory requirements. The study recommends that insurance companies integrate regulatory compliance considerations into appraisal design, automate appraisal processes, strengthen feedback mechanisms, and that Insurance Regulatory Authority provide clearer guidance on how appraisal and conduct standards should interact.

Keywords: performance appraisal, industry regulation, organizational performance, moderating effect, insurance companies, Nairobi city

DOI https://doi.org/10.33803/5cs5yePublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/5cs5ye ↗

36.Portfolio Diversification Practices and Financial Performance of Registered Fund Managers in KenyaAbstract

Patrick Kamau Kahura^1*

Portfolio diversification is an important investment management practice for reducing investment risk and improving financial outcomes. This study examined the effect of portfolio diversification practices on the financial performance of registered fund managers in Kenya. The study was motivated by the continued expansion of Kenya’s fund-management industry and the increasing volume of assets managed by fund managers, which heightened the need for effective investment allocation and risk management. Anchored on Modern Portfolio Theory, the study adopted a positivist research philosophy and a descriptive and correlational research design. A quantitative approach was employed, with data collected using structured questionnaires from respondents drawn from registered fund managers in Kenya. A total of 180 questionnaires were issued, of which 160 were successfully returned, yielding a response rate of 88.9%. Data analysis was done using descriptive and inferential statistics, including frequencies, percentages, means, standard deviations, Pearson correlation and simple linear regression. Hypothesis testing was conducted at the 5% significance level. Descriptive findings indicated that portfolio diversification practices were generally adopted, with an overall mean of 3.90 (SD = 1.12). The inclusion of a wide range of financial instruments recorded the highest mean (M = 4.01, SD = 1.10), while regular monitoring and control of portfolio concentration recorded the lowest (M = 3.73, SD = 1.12). Correlation analysis established a strong positive and statistically significant relationship between portfolio diversification and financial performance (r = 0.756, p < 0.001). Regression results showed that portfolio diversification explained 57.2% of the variation in financial performance (R² = 0.572), with the model being statistically significant (F(1,158) = 210.759, p < 0.001). The regression coefficient confirmed a positive and significant effect (β = 0.756, t = 14.473, p < 0.001), leading to rejection of the null hypothesis. The study concluded that portfolio diversification significantly and positively influenced financial performance.

Keywords: portfolio diversification, financial performance, fund managers, investment management, asset allocation, Kenya

DOI https://doi.org/10.33803/3hp2ssPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/3hp2ss ↗

37.Product Development Strategy and Competitive Advantage of Farm Input Distribution Companies in Nakuru County, KenyaAbstract

Kiplangat Gideon Ngetich^1*, Ronald CHEPKILOT¹, Njenga GITAHI^1*

Farm input distribution companies are important to agricultural productivity and food security in Kenya, yet their competitiveness is challenged by limited product innovation, customer retention, and market responsiveness. This study examined the relationship between product development strategy and competitive advantage among farm input distribution companies in Nakuru County, Kenya. The study was anchored in Dynamic Capabilities Theory and adopted a correlational research design. The target population comprised 238 registered farm input distribution companies in Nakuru County. A census approach was applied, with 24 firms in Naivasha Sub-County participating in the pilot study and 214 firms included in the main study. The firm was the unit of analysis, while one senior manager served as the unit of observation. A total of 160 questionnaires were completed, yielding a response rate of 74.8%. Data were analysed using descriptive statistics, Pearson correlation, and simple linear regression in SPSS Version 27. The findings indicated a strong positive and statistically significant relationship between product development strategy and competitive advantage (r = 0.831, p < 0.001). Regression results showed that product development strategy explained 69.1% of the variance in competitive advantage (R² = 0.691). The model was statistically significant (F = 353.188, p < 0.001), and the regression coefficient was positive and significant (β = 0.828, t = 18.793, p < 0.001). Given the correlational design, these findings indicate association rather than causation. The study concludes that strengthening research and development, customer-driven product innovation, and process improvement is associated with competitive advantage. It contributes to Dynamic Capabilities Theory by demonstrating its relevance to agricultural distribution through innovation and adaptation. The study recommends that distributors prioritize market-responsive product development and process improvement, while policymakers and regulators support innovation, quality assurance, and efficient input distribution to strengthen long-term sectoral competitiveness and contribute to Kenya’s food security.

Keywords: product development strategy, competitive advantage, farm input distribution companies, growth strategies, innovation

DOI https://doi.org/10.33803/m2zreuPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/m2zreu ↗

38.Project Planning and Performance of Road Construction Projects: Evidence from Kenya National Highways Authority in Kenya’s South Rift RegionAbstract

Eric Muathe Musyoka^1*, Richard Nyaoga², Jeptepkeny Chebet Bowen¹

The study conceptualized project initiation, project planning, project execution and project monitoring and control as the independent variables, while project performance was the dependent variable. Project performance was operationalized as a multidimensional construct. Pearson correlation and regression analysis were used to establish the individual and combined relationship between the four independent variables and project performance. The study was motivated by persistent challenges in road construction project delivery, including delays, cost overruns and failure to achieve planned outputs. A positivist research philosophy and a descriptive and correlational research design were adopted. The study employed a quantitative approach, with primary data collected using structured questionnaires. Data were analysed using descriptive statistics, Pearson correlation and regression analysis. Descriptive findings indicated that project planning practices had been well implemented, with a mean score of 3.90 (SD = 1.12). Risk assessment before implementation recorded the highest mean (M = 4.02), while documentation and sharing of implementation plans recorded the lowest mean (M = 3.68). Inferential findings established a strong positive and statistically significant relationship between project planning and project performance (r = 0.720, p < 0.001). The regression model indicated that project planning explained 51.8% of the variation in project performance (R² = 0.518). The analysis of variance established that the regression model was statistically significant, F(1,115) = 123.55, p < 0.001. Regression results showed that project planning had a positive and statistically significant influence on project performance (β = 0.720, t = 11.116, p < 0.001). The null hypothesis that project planning had no statistically significant influence on project performance was rejected. The study concluded that effective project planning contributed to improved performance of road construction projects. The study recommended that KeNHA strengthen project planning practices through realistic scheduling, effective resource allocation, comprehensive risk assessment, regular review of plans and enhanced stakeholder coordination.

Keywords: project planning, project performance, road construction projects, KeNHA, infrastructure development

DOI https://doi.org/10.33803/cyawptPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/cyawpt ↗

39.Risk Management Practices and Financial Performance of Microfinance Institutions in Nairobi County, Kenya: Moderating Role of Regulatory ComplianceAbstract

Abdirahim Mohamud Mudo^1*, Stephen Oloo Magadi², Kiplimo Sirma²

Risk identification practices are increasingly important in the microfinance sector, where institutions operate amid credit, liquidity, operational, market, and emerging risks. Effective risk identification enables institutions to take corrective measures, minimize losses, and sustain financial performance. This study examined the effect of risk identification practices on the financial performance of microfinance institutions in Nairobi County, Kenya. Guided by Enterprise Risk Management Theory, the study adopted a positivist philosophy, descriptive and correlational research design, and quantitative approach. The target population comprised managers, risk officers, credit officers, and finance officers in licensed microfinance institutions. The study collected primary data using structured five-point Likert-scale questionnaires. Of 165 questionnaires issued, 144 were returned, representing an 87.3% response rate. Data were analyzed in SPSS using descriptive statistics, Pearson correlation, and simple linear regression. Risk identification practices were generally implemented (M = 3.95, SD = 1.15), with risk prioritization recording the highest mean (M = 4.06) and documented risk mapping the lowest (M = 3.74). Risk identification practices were strongly and positively related to financial performance (r = 0.638, p < 0.01). Regression results showed that risk identification explained 40.7% of the variation in financial performance (R² = 0.407), with a positive and significant effect (β = 0.218, t = 2.907, p = 0.004). The model was statistically significant, F (1, 142) = 97.21, p < 0.001. A pilot study of 10 respondents across two microfinance institutions in Kiambu County showed satisfactory reliability, with Cronbach’s alpha coefficients ranging from 0.812 to 0.879 and an overall alpha of 0.861. The study concluded that risk identification practices significantly influenced financial performance. It recommends strengthening risk mapping, early-warning systems, employee risk awareness, and identification and prioritization of emerging risks, while maintaining systematic documentation of identified risks across institutional functions and decision-making processes at all levels.

Keywords: risk identification practices, financial performance, microfinance institutions, risk management, Nairobi county, Kenya

DOI https://doi.org/10.33803/hp8kakPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/hp8kak ↗

40.Risk Mitigation Strategies and Financial Performance of General Insurance Companies in Nairobi County, KenyaAbstract

Phoebe Aluoch Olooh^1*, Patrick Kibati², Stella Cheraisi Korir²

Risk management is important within the insurance industry because insurance companies operate in an environment characterized by underwriting risks, claims volatility, operational disruptions, fraud, market uncertainty and emerging risks. Effective risk mitigation strategies are essential for reducing losses, protecting policyholders and maintaining financial performance. This study evaluated the effect of risk mitigation strategy on the financial performance of general insurance companies in Nairobi County, Kenya. The study was guided by Enterprise Risk Management Theory, which emphasizes identification, assessment, response and monitoring of risks. A positivist research philosophy, descriptive and correlational research design, and quantitative research approach were adopted. A total of 225 questionnaires were issued, of which 187 were returned, representing an 83.0% response rate. Data were collected using structured questionnaires. Descriptive analysis involved means and standard deviations, while Pearson correlation and simple linear regression established the relationship and effect of risk mitigation strategy on financial performance. The descriptive findings indicated that risk mitigation strategies were implemented, with an overall mean of 3.70 (SD = 1.28). Consistent adherence to underwriting guidelines recorded the highest mean (M = 3.89, SD = 1.24), while maintaining reinsurance arrangements recorded the lowest mean (M = 3.48, SD = 1.30). Correlation analysis established a strong positive and statistically significant relationship between risk mitigation strategy and financial performance (r = 0.786, p < 0.01). Regression analysis showed that risk mitigation strategy explained 61.8% of the variation in financial performance (R² = 0.618). The regression model was statistically significant (F(1,185) = 299.293, p < 0.001), while the regression coefficient confirmed a positive and statistically significant effect (β = 0.786, t = 17.300, p < 0.001). The null hypothesis was rejected. The study concluded that effective risk mitigation strategies significantly influenced financial performance and recommended strengthened regulatory oversight, enhanced underwriting controls, regular risk-management reviews and improved reinsurance arrangements.

Keywords: risk mitigation strategy, financial performance, general insurance companies, risk management, underwriting, reinsurance, Kenya

DOI https://doi.org/10.33803/u3kkntPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/u3kknt ↗

41.The Effect of Operational Practices On Employee Mental Health In Mission Hospitals In South Rift, KenyaAbstract

Nicholas Kipyegon Mutai^1*

Healthcare workers in mission hospitals carry a mental health burden that remains largely undocumented. WHO (2017) reports that about 10% of Kenya's workforce loses an average of 36 working days annually due to depression, and the Ministry of Health (2023) estimates 3.7 million Kenyan workers currently live with a mental health condition. This study examined the effect of workload management practices on employee mental health in mission hospitals in South Rift, Kenya, anchored on the Job Demand-Resource (JD-R) Theory, which identifies workload practices that either drive or prevent stress and burnout among hospital workers. Quantitative data came from structured questionnaires administered to 338 clinical and non-clinical employees, while key informant interviews with hospital administrators and departmental heads provided qualitative insight. The sample was fairly balanced (45% male, 55% female), with 72% of employees under 45 years old, the largest group aged 25–34. Descriptive statistics showed employees frequently multi-tasked (mean=4.040), felt overwhelmed by their duties (mean=3.94), and worked overtime regularly (mean=3.88), while staffing levels were disproportionate to patient volumes (mean=2.02). Work-life balance practices had a strong, statistically significant positive effect on mental health (B=0.800, β=0.520, p=0.003), with a strong correlation (r=0.757, p<0.001). Qualitative findings confirmed that staff shortages and emergency demands continued to strain employees' well-being. The study concludes that workload management practices significantly influence employee mental health in these hospitals—equitable workload distribution, manageable overtime, adequate staffing, and sufficient breaks can improve well-being. It recommends that mission hospitals in South Rift align staffing with patient volumes and workload demands, and that management conduct regular workload assessments.

Keywords: workload management practices, employee mental health, mission hospitals

DOI https://doi.org/10.33803/3zmh5jPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/3zmh5j ↗

42.Computational Investigation of How Spatial Myosin Motor Clustering Regulates Actin Filament Mechanics and Motor EngagementFull paper

Morris Muriithi Wachira^1*⁠iD, Douglas Kagoiya Ng'ang'a²⁠iD, Takahiro Nitta²⁠iD, Samuel Macharia Kang'iri^1*⁠iD

In in vitro motility assays, myosin motors are conventionally modeled as uniformly distributed across the substrate. In practice, both engineered assay surfaces and cellular actomyosin networks present motors in discrete spatial clusters, and how this organization affects filament mechanics and motor engagement remains poorly understood. Here we develop a two-dimensional Brownian dynamics model of actin filament transport over a spatially patterned motor field. Using a full 3×3×3 factorial design, we independently varied cluster density, cluster radius, and clustering fraction while holding the total motor number constant, and combined these three parameters into a single clustering index representing overall clustering intensity to permit comparison across geometries. Filament mechanics were quantified by the persistence length of the filament contour, and motor engagement dynamics by the coefficient of variation of the bound-motor count. Mean persistence length increased primarily with cluster radius, and depended on clustering fraction only through its interaction with cluster density. Bound-motor fluctuations behaved differently, increasing strongly with both cluster density and cluster radius, and were most sensitive to clustering fraction at intermediate cluster densities. The clustering index captured the overall rise in both responses with clustering intensity, but geometries sharing similar index values produced measurably different mechanical and kinetic behavior, showing that a single scalar descriptor does not fully collapse the response surface. Spatial motor organization therefore governs actin filament transport through two separable pathways: a mechanical pathway set by the length scale over which motors engage the filament, and a kinetic pathway set by the statistics of motor binding and release. Motor geometry should accordingly be represented explicitly when modeling actomyosin systems, and treated as a design variable when engineering biomolecular transport platforms.

Keywords: actomyosin, in vitro motility assay, Brownian dynamics simulation, molecular motor clustering, filament persistence length

DOI https://doi.org/10.33803/6gsbjhPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/6gsbjh ↗

43.Investigating the Impact of Strict Laws on Corruption: A Mathematical Modelling ApproachAbstract

David Musyoka Kinyili^1*, Dominic Makaa Kitavi²

This paper develops a deterministic model that addresses the dynamics of corruption in the Kenyan context. Despite several researchers making effort to develop corruption models, the existing models suffer the shortcomings such as: Failure to capture the complexity of corruption, failure to incorporate strict laws against corruption and adherence to the laws, failure to factor in moral behavior that act against, and failure to fit the models to real data. Therefore, the current study comes in handy to address this gap by developing a mathematical model that addresses the dynamics of corruption and carefully incorporates all the aforementioned limitations of the existing models establishing its novelty. The motivation of the study is that, although anti-corruption bodies with laws have been established, corruption still continues to be deep-rooted in many world regions including Kenya. We compute the corruption reproduction number, denoted by R_0^C. We find that the model possesses a unique corruption-free and a positive corruption-endemic equilibria, locally and globally asymptotically stable whenever R_0^C<1 and R_0^C>1 respectively. The developed model is fitted to real data of corruption reported in Kenya for realism and accurate parameter estimation used for numerical results. Sensitivity analysis results establish that, serious implementation of strict laws, adherence to the laws, and honest behavior substantially reduce the corruption reproduction number to less that unity, eradicating corruption. Simulations indicate that thorough implementation of strict laws against corruption can eradicate corruption in a relatively short time. Further, the aspect of obedience to the laws and the behavioral practice of trust and honesty contribute significantly to eliminating corruption. The findings obtained inform policy on anti-corruption agencies to formulate/strengthen and implement/enforce stringent laws against corruption in order to curtail it, create serious public awareness on corruption and involve the church in instilling the qualities of trust and honesty to eliminate corruption.

Keywords: mathematical modeling, corruption-free-equilibrium, corruption-reproduction-number, corruption-endemic-equilibrium, model calibration, simulations

DOI https://doi.org/10.33803/xrfrm5Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/xrfrm5 ↗

44.An Agent Based Model for Social Engineering Resilience Based on User BehaviourAbstract

Bryan Ruiru^1*, Nelson Masese¹, Andrew Kipkebut¹

Vishing (voice phishing) and smishing (SMS phishing) are increasingly common in Kenya, where high mobile-phone and mobile-money usage creates attractive targets for attackers who exploit human psychology rather than technical vulnerabilities. Current defenses often overlook behavioral, psychological, and contextual factors that shape user responses, highlighting the need for a behavior-centered modeling framework. This study developed and evaluated an agent-based model (ABM) of social-engineering resilience based on user behavior in Kenya. A mixed-method approach combined descriptive analysis of secondary reports, literature-based parameterization, and expert-informed validation. Behavioral variables, including awareness, trust, susceptibility, risk perception, and reporting tendency, were derived from recurring constructs identified in the literature and represented as agent attributes and decision-rule weights. A NetLogo ABM simulated heterogeneous users and attackers under three scenarios. The baseline scenario had no intervention. Moderate intervention combined limited awareness training, verification, and reporting support, while strong intervention applied higher-intensity training, verification, and reporting measures with social-network reinforcement. In this model, social-network reinforcement refers to the influence of reports and protective behavior shared among connected users. High-risk users were those classified by the model as having high susceptibility and low protective behavior. Layered awareness campaigns combined repeated training, warnings, and reporting guidance. The model showed a strong reduction in attack success as interventions intensified. Cumulative attack success was approximately 51.8% in the baseline, 14.3% under moderate intervention, and 6.5% under strong intervention. Rolling 60-day success rates stabilized at about 54.5%, 8.4%, and 1.7%, respectively. Sensitivity analysis and expert review indicated that the simulated patterns were plausible and consistent with known trends. Agent-based modelling offers a credible and flexible tool for simulating social-engineering resilience in Kenya’s mobile-centric environment. The model enables policymakers and cybersecurity practitioners to test intervention strategies before real-world deployment.

Keywords: agent-based model, cybersecurity resilience, simulation, smishing, social engineering, vishing

DOI https://doi.org/10.33803/29pej2Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/29pej2 ↗

45.Cybersecurity Controls and Operational Performance of Private Healthcare Centres in Nakuru City, KenyaAbstract

Duncan Mutai^1*

The increasing adoption of digital technologies in healthcare has improved information management, patient record management, and service delivery while increasing exposure to cybersecurity threats. This study analysed the influence of cybersecurity controls on the operational performance of private healthcare centres in Nakuru City, Kenya. The study was guided by Protection Motivation Theory and adopted a positivist philosophy, descriptive and correlational research design, and a quantitative approach. Data were collected using structured questionnaires from personnel in private healthcare centres. Out of 156 questionnaires issued, 136 were returned, representing an 87.2% response rate. Data were analysed using descriptive statistics, Pearson correlation, and linear regression. Descriptive findings indicated that cybersecurity controls were relatively well implemented, with an overall mean of 3.86 (SD = 1.14). Monitoring antivirus alerts and logs to detect and respond to potential threats promptly recorded the highest mean (M = 4.08, SD = 1.06). Correlation analysis established a strong positive and statistically significant relationship between cybersecurity controls and operational performance (r = 0.684, p < 0.01). Regression analysis showed that cybersecurity controls explained 46.8% of the variation in operational performance (R² = 0.468). The regression model was statistically significant, F (1, 134) = 124.03, p < 0.001, while cybersecurity controls had a positive and significant influence on operational performance (β = 0.684, t = 11.210, p < 0.001). The null hypothesis was therefore rejected. The study concluded that stronger cybersecurity controls were associated with improved operational performance among private healthcare centers. It recommended that healthcare managers strengthen cybersecurity infrastructure, authentication, encryption, security monitoring, backup and recovery systems, and employee cybersecurity awareness. Policymakers and regulatory agencies were also recommended to strengthen cybersecurity standards and compliance monitoring within private healthcare facilities. The study demonstrated that effective cybersecurity controls were an important organizational capability for supporting efficient, reliable, and continuous healthcare service delivery.

Keywords: cybersecurity controls, operational performance, private healthcare centres, healthcare information systems, data protection

DOI https://doi.org/10.33803/a5kewfPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/a5kewf ↗

46.Mobile Phone Number Analysis as Digital Forensics Tool for Detection and Investigation of Cybercrime in Kenya and East AfricaAbstract

Julius Kinuthia Njiraini^1*, Jane Kuria², Benson Kituku³, John Njiraini⁴

Mobile phone-based cybercrimes is one of the rising form of cybercrimes in Kenya and East Africa, however very few study have been done to understand how it is being detected and analyzed. The study aims to improve the efficiency of investigations by helping in the correlation of evidence in various digital platforms as well as in attributing criminal activities to suspects through improving the efficiency of investigations by attributing criminal activities to suspects. The research focused on the most prevalent cybercrimes that have been committed using mobile phone numbers in Kenya and East Africa, as well as the importance of phone number forensics in detecting and investigating cybercrimes. The study employed quantitative approach of research whereby the participants of the study consisted of police officers from the Cybercrime unit of the Directorate of Criminal Investigation (DCI). The collection of quantitative data was done through structured questionnaire. The results played an important role in investigating various forms of cybercrime. The analysis of mobile phone numbers was noted to be one of the most important methods of digital forensics in the fight against cybercrimes due to the wide use of mobile communication technologies. The integration of the phone number analysis method into traditional techniques used in digital forensics improved the ability of law enforcement officers to track down, investigate, and prosecute cyber criminals.

Keywords: Cybercrime, mobile phone analysis, digital forensics, mobile money fraud, digital evidence, DCI, cyber investigations, phone number tracking

DOI https://doi.org/10.33803/ff4wmePublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ff4wme ↗

47.Characterizing the Impact of Land Use Changes, Climate Variation, and Anthropogenic Factors on Tea Production in Kangema Sub‑County, Murang'a CountyPoster

Mary Njeri Mwangi^1*, Mark Kipkurwa Boitt^1*, Charles Ndegwa Mundia^1*

Background. Tea production is influenced by a complex interaction of environmental and human factors. In Kangema Sub-County, changes in land use, climate variability, and anthropogenic activities affect the availability and suitability of land for tea cultivation and consequently influence production. However, the combined contribution of these factors to changes in tea production is not well characterized. This study therefore investigates the relationship between land-use change, climate variables, anthropogenic factors, and tea production in Kangema Sub-County, Kenya, providing insights into the drivers of production change and informing sustainable tea-production planning.

Methods. Where: Kangema Sub-County, Murang'a County, Kenya. On what: Tea production and its drivers — land use, climate (rainfall, temperature), and anthropogenic factors — 1993–2023. How: Land use: classified Landsat imagery (1994, 2003, 2013, 2023) into tea, vegetation, bare land, built-up/roads (ha). Climate: analyzed rainfall (mm) and temperature (°C) records and trends. Anthropogenic: survey data on tea growers and inputs. Analysis: computed changes over time and compared variables against tea production.

Results. Land use (1994–2023): Tea plantation 2,954 → 3,762 ha (+27%); Vegetation 10,094 → 7,483 ha (−26%); Bare land 3,277 → 3,114 ha (−5%); Built-up & roads 1,351 → 3,317 ha (+146%). Climate (1993–2023): Rainfall 1,140 → 1,555 mm (+36%); Temperature 16.2 → 17.0 °C (+0.8 °C). Anthropogenic (1993–2023): Tea growers 7,306 → 10,381 (+42%). Packaging waste (1993–2025): Waste 483 → 195 g/kg; Recycling 12% → 42%.

Conclusions. Land-use change and anthropogenic factors influenced tea production in more strongly than the climate variables assessed. Vegetation declined by 26%, built-up and road area expanded by 146%, and tea plantations grew by 27%, 1993–2023. Tea growers increased by 42%, rainfall rose from 1,140 mm to 1,555 mm, and temperature rose from 16.2 °C to 17.0 °C. Land-use and human factors, not climate alone, are the primary drivers of change in tea production.

Keywords: tea production, land-use change, climate variation, anthropogenic factors

DOI https://doi.org/10.33803/bbsnq4Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/bbsnq4 ↗

48.Characterizing Urban Frontiers Dynamics Across Kenyan CitiesAbstract

Vinither Kemunto Nyabuto^1*, Daniel Orongo Nyangweso¹, Duncan Maina Kimwatu¹

As urban population grows, urban areas tend to stretch and extend into the surrounding land use, creating urban frontiers. These areas are considered the leading edge of urban growth and are characterized by intense land-use changes, yet they remain a critical blind spot. To address this, this study quantified and compared the spatial structure and temporal dynamics of urban frontiers in Nairobi Metropolitan, Mombasa and Eldoret using land use land cover analysis and spatial metrics. Landsat imagery for 1990, 2000, 2010 & 2020 was classified using artificial neural network, and built-up areas were integrated with nightlight data to delineate and map urban frontiers. Spatial metrics defining pace, shape and direction of urban growth within urban frontiers were standardized and combined into composite indices. Nearest neighbour analysis showed significant clustering of urban frontiers in Nairobi across all periods (p<0.001) and in Mombasa city during 1990-2000 and 2010-2020 (p<0.05) while Eldoret exhibited a regular, non-clustered arrangement of urban frontiers (p>0.1). PERMANOVA analysis showed that both time period (p=0.001) and study area (p=0.001) had a strong influence on the characteristics of urban frontiers. Additionally, Significant two-way interactions and three way interaction between time period, study area and spatial orientation (p <0.05) proved that urban frontier dynamics vary depending on the time period, study area and zone they occur.The stratified pairwise test revealed that urban frontiers characteristics within Nairobi city and Mombasa city in 1990-2000 differed significantly from both subsequent periods while Eldoret city showed no significant temporal variations. Based on these findings, we can conclude that the spatial structure of urban frontiers is heterogeneous across Kenyan cities and varies with time. It is therefore important to incorporate urban frontiers in urban development and implementation plans that are localized to meet the unique needs of different cities in Kenya.

Keywords: urban frontier, urbanization, spatial metrics, Kenyan cities, urban frontier dynamics, urban frontier spatial structure

DOI https://doi.org/10.33803/3n6ktmPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/3n6ktm ↗

49.Evaluating the Readiness of Kenya’s Land Administration System for Real-Time Cadastral Updating and Spatial ReferencingAbstract

Jeremiah Omeke^1*, Mark Kipkurwa Boitt¹, Daniel Orongo Nyangweso¹

Several digital land administration initiatives, including Ardhisasa, have been implemented in Kenya to improve land governance. Nevertheless, the readiness of the current system for real-time cadastral updating remains unclear. This study evaluated Kenya's land administration system through a document review, stakeholder surveys, benchmark analysis, and field observations, guided by the Land Administration Domain Model (LADM), Fit-for-Purpose Land Administration, Cadastre 2014, and the Social Tenure Domain Model (STDM). A structured questionnaire was administered using purposive sampling to 417 land administration professionals and stakeholders across Kenya, and responses were analyzed using descriptive statistics, Pearson correlation, and Cronbach's Alpha reliability testing alongside a qualitative document analysis of the Land Act, the Land Registration Act, and the National Land Policy (2023) using ATLAS.ti. The gap this study addresses is the absence of an empirical account of how ready Kenya's land administration system is for continuous, real-time cadastral updating and reliable spatial referencing, despite ongoing digitization efforts. Findings show significant limitations in cadastral updating, interoperability, spatial referencing, and institutional coordination, even though digital transformation has improved the accessibility of land information. The system showed moderate compliance with international standards for legal interoperability and real-time updating, and 96.3% of respondents rated the temporal (4D) dimension of land administration as important or very important, while 77.9% supported integrating a geographic gazetteer for spatial referencing. These findings indicate that Kenya's land administration system provides a workable foundation but requires the integration of Earth Observation technologies, machine learning, and dynamic cadastral models to support real-time land administration. The results point to an incremental reform pathway, extending the existing 2D parcel system with a linked 3D object layer, as more immediately achievable than a full 3D cadastral overhaul, and offer evidence to guide policymakers on prioritizing legal, institutional, and technical investments for real-time land administration in Kenya.

Keywords: land administration, cadastral updating, spatial referencing, earth observation, machine learning, Kenya

DOI https://doi.org/10.33803/yyqrpqPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/yyqrpq ↗

50.Fine-Grained Tree Species Identification in Heterogenous Ecosystems Using Integrated Aerial and Terrestrial ImagesAbstract

Leonard Sanya^1*, Juliet Moso², Ciira Maina³

Accurate fine-grained tree species identification is essential for effective biodiversity monitoring and forest conservation. While Unmanned Aerial Vehicle (UAV) imagery offers high-resolution spatial coverage, aerial-only classification in highly diverse ecosystems is often constrained by crown occlusion, shadows, and inter-class morphological similarities. Conversely, terrestrial photography captures discriminative ground-level traits such as bark texture and leaf arrangement but lacks broader spatial context. The primary contribution of this study is the introduction and analysis of the Miti360 dataset, collected from a 770-hectare reforested stand in Kieni Forest located in Kiambu County, Kenya. The dataset comprises 1,208 individual trees (601 collected in 2024 and 607 in 2025) spanning 26 distinct species. The study utilizes precise GPS locations to spatially associate terrestrial photographs with their corresponding UAV crops. Furthermore, to prevent geographic data leakage, these GPS locations were used to split the dataset into spatial blocks, ensuring zero overlap between the training, validation, and testing sets. We implemented a dual-stream deep learning architecture that uses a Vision Transformer (ViT) to encode global canopy structures and a Convolutional Neural Network (CNN) to extract localized terrestrial features, which are integrated through a cross-attention mechanism. During evaluation, the model was tested using aerial features only, terrestrial features only, and the combined aerial--terrestrial features. Using aerial features alone yielded an accuracy of 26.76% and an F1-score of 15.47%, while terrestrial features alone achieved 80.28% accuracy and a 78.49% F1-score. Combining both feature sets through cross-attention produced the best performance, with an overall accuracy of 85.92% and an F1-score of 85.47%. These findings demonstrate the value of complementary aerial and terrestrial information for tree species identification in complex forest environments. Future work will include continuous model fine-tuning and comparative analysis with the unimodal models trained on aerial and terrestrial features only.

Keywords: forests, remote sensing, cross-attention, tree species classification

DOI https://doi.org/10.33803/76vn62Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/76vn62 ↗

51.Geology and Copper Ore Mineralization in the Mozambique Belt Rocks of the Kwa Vonza–Mutonga Area, Kitui West, KenyaAbstract

Stellah Syombua Kitonga^1*⁠iD, Patrick Chege Kariuki², Noah Misati Kerandi¹

The study area in western Kitui County, southeastern Kenya, is underlain by Neoproterozoic rocks of the Mozambique Belt, characterized by metamorphism, deformation, granitoid intrusions and mineralization. Despite its mineral potential, detailed studies of copper mineralization in the Kwa-Vonza–Mutonga area remain limited. This study investigated the geology and copper mineralization, focusing on host lithologies, structural controls and copper enrichment. An integrated approach involving Sentinel-2 Level-2A remote sensing, geological field mapping, petrography, X-ray fluorescence (XRF) and X-ray diffraction (XRD) was applied. Sentinel-2 imagery was processed in ArcGIS 10.8 using false-colour composites, spectral band ratios and image enhancement techniques to support lithological discrimination and identify alteration zones and structural lineaments. Enhanced imagery, band-ratio maps, alteration maps and lineament information were integrated with field observations to improve geological interpretation and delineation of mineralized zones. Lithological units were established primarily through field mapping based on mineralogical, textural and structural characteristics and validated using petrographic analysis, with remote sensing providing complementary spatial information. The study established that the area comprises granitoid gneisses, banded biotite gneisses, migmatites, meta-mafic and meta-felsic rocks, granitic and schistose gneiss, granulite and surficial deposits. Copper mineralization is predominantly associated with mafic and felsic rocks, particularly metagabbro and metagranite, and is concentrated along deformed contacts between mafic and felsic units, faults, fractures, joints, foliation planes and vugs. Remote sensing identified lineaments and alteration-related anomalies coinciding with mapped mineralized zones and artisanal workings, while geochemical results confirmed copper enrichment in altered and structurally controlled lithologies. A preliminary assessment indicates approximately 466 tonnes of contained copper within the mineralized zone. This estimate is based on geological observations from one trench, measured dimensions, bulk-density determination and available geochemical data and should not be regarded as a formal Mineral Resource for the whole study area. Drilling and subsurface investigations are required for resource validation.

Keywords: copper mineralization, Mozambique belt, GIS, remote sensing, structural controls, geochemistry

DOI https://doi.org/10.33803/7vqystPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/7vqyst ↗

52.Integrated Earth Observation and Machine Learning Modelling of Subsurface Soil Moisture in a Heterogeneous Vegetated Landscape: The Nzoia River Basin, Western KenyaAbstract

Elizabeth Ayongo Okumu^1*

Soil moisture governs key processes within the hydrological cycle, shaping runoff generation, evapotranspiration, water availability for vegetation, drought dynamics, and groundwater recharge. Despite its importance, subsurface soil moisture in tropical, spatially variable landscapes remains inadequately understood, largely due to sparse field monitoring networks and a research emphasis that has historically favoured surface-layer moisture estimation over deeper soil horizons. This research presents an integrated framework combining Earth Observation data with machine learning to model subsurface soil moisture patterns across the Mumias–Nzoia Basin in Western Kenya. A range of data sources – Sentinel-1 radar imagery, Sentinel-2 optical data, terrain elevation models, ERA5-Land climate reanalysis, and SoilGrids soil property data – were combined with field-based measurements to assess how soil moisture varies across space and time between 2017 and 2025. Predictor variables spanning vegetation cover, topography, climatic conditions, and soil properties were fed into Random Forest and Extreme Gradient Boosting algorithms, and a stacked ensemble model was built to enhance overall prediction accuracy. Initial results indicate high model performance: the optimized Random Forest model reached an R² of 0.9653, RMSE of 0.0442, MAE of 0.0297, and NSE of 0.9653, whereas the optimized XGBoost model performed slightly better, with an R² of 0.9718, RMSE of 0.0399, MAE of 0.0275, and NSE of 0.9718. The stacked ensemble achieved a similar level of accuracy, with an R² and NSE of 0.9711. This modelling framework lays the groundwork for producing continuous, spatially and temporally resolved subsurface soil moisture maps spanning 2017 to 2025. The findings highlight the value of integrating satellite-derived data with machine learning to enhance subsurface soil moisture assessment in mixed agricultural and vegetated environments, offering practical support for water resource planning, agricultural decision-making, drought early-warning systems, and climate adaptation efforts in regions of Kenya where ground-based data remain limited.

Keywords: subsurface soil moisture, earth observation, machine learning, random forest, XGBoost, stacked ensemble modelling

DOI https://doi.org/10.33803/3g9vvhPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/3g9vvh ↗

53.Integrating Earth Observation and Hydrogeospatial Data for Groundwater Inundation Risk Assessment Using ANFISAbstract

Grace Bulenywa Isomi^1*, Moses Murimi Ngigi^1*, Daniel Orongo Nyangweso^1*

Groundwater inundation occurs when rising groundwater levels cause water to accumulate at or near the land surface, posing risks to agriculture, infrastructure, ecosystems, and communities, particularly within low-lying floodplains. This study integrates Earth Observation and hydrogeospatial data with an Adaptive Neuro-Fuzzy Inference System (ANFIS) to assess groundwater inundation risk in the Lower Nyando River Basin, Kenya. Multi-source geospatial datasets were used to characterize environmental controls on groundwater inundation, including groundwater depth, elevation, hydraulic conductivity, recharge, precipitation, temperature, Normalized Difference Vegetation Index(NDVI), Normalized Difference Water Index(NDWI), Topographic Wetness Index (TWI), Height Above Nearest Drainage (HAND), Vertical Distance to Channel Network (VCDN), lineament density, land use/land cover, soil type, slope, Terrain Ruggedness Index (TRI), and Horizontal Overland Flow Distance (HOFD). A Groundwater Inundation Index was developed from normalized predictor variables to represent spatial variations in inundation risk. The ANFIS model employed Gaussian membership functions, with 70% of observations used for training and 30% for independent validation. The model achieved an R² of 0.877 and RMSE of 105.47 m during training, while validation produced an R² of 0.859 and RMSE of 113.84 m. The resulting classification achieved a ROC-AUC of 0.953, indicating strong discrimination between areas of different groundwater inundation risk. Groundwater depth and HAND emerged as the most influential predictors. The findings demonstrate that combining Earth Observation with hydrogeospatial variables and ANFIS provides an effective framework for identifying groundwater inundation-prone areas and can support floodplain planning, agricultural management, and climate-resilient water-resource management in data-limited regions.

Keywords: earth observation, hydrogeospatial data, ANFIS, groundwater inundation, risk assessment, remote sensing, GIS

DOI https://doi.org/10.33803/feqd4aPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/feqd4a ↗

54.Machine Learning Modeling of Multivariate Socio-Ecological Drivers of Maize Crop Health: A Case Study of Siaya County, KenyaPoster

Tom Omollo Alando^1*, Duncan Maina Kimwatu¹

The health of maize crops serves as a critical indicator of food security and socio-economic stability, especially in agriculture-dependent regions like Siaya County. Over recent decades, maize productivity has faced growing threats from the interplay of social, ecological, and climatic pressures, such as variable rainfall, extreme temperatures, changing land use, and soil moisture stress. Although these factors are individually important, they are frequently examined in isolation, which restricts a holistic understanding of their combined effects on maize crop health. This study fills that gap by developing a multivariate social-ecological index and assessing its impact on maize crop health in Siaya County, Kenya, drawing on multi-source geospatial data from USGS Landsat imagery, CHIRPS precipitation, ISRIC SoilGrids, WorldPop datasets covering 1995–2025. The integrated datasets included satellite-derived vegetation indices (NDVI and VHI), soil moisture index and land use/land cover; temperature condition indices derived from Landsat products, topographic variables from the SRTM digital elevation model; soil properties and population density. Random Forest and Extreme Gradient Boosting algorithms were used to capture nonlinear relationships between environmental drivers and maize crop health, while logistic regression examined the link between the composite index and crop health outcomes. The results showed marked spatial and temporal variation in maize crop health across the county. Ecological and climatic factors together shaped crop health outcomes, with soil moisture as aleading driver (about 41%). Climaticvariables, particularly precipitation and temperature variability; also contributed notably (roughly 13–15%), while land use and population factors had a moderate effect. Random Forest slightly outperformed Extreme Gradient Boosting in predictive accuracy (R² = 0.70 versus 0.68), and logistic regression confirmed a strong association between the index and crop health (AUC = 0.93–0.99). These findings show that combining social,ecological, and climatic variables through machine learning offers a reliable approach for monitoring maize crop health and informing climate-resilient agricultural planning.

Keywords: maize crop health, social-ecological index, machine learning, random forest, extreme gradient boosting

DOI https://doi.org/10.33803/3t3kr3Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/3t3kr3 ↗

55.Satellite Image Super-resolution Using Diffusion Based ApproachAbstract

James Muiru^1*, Ciira Maina¹⁠iD, Cedric Kiplimo¹

The trade-off between spatial and temporal resolution in satellite imagery limits the use of available data for fine-scale Earth observation. Although Sentinel-2 provides frequent coverage, its 10 m spatial resolution can be insufficient for agricultural monitoring and environmental assessment. This study investigates a pixel-space diffusion-based super-resolution approach using Super-Resolution via Iterative Refinement (SR3) to generate 2.5 m spatial-resolution estimates from 10 m Sentinel-2 multispectral imagery. Unlike latent diffusion approaches, SR3 performs diffusion and denoising in image pixel space without transforming imagery into a latent representation. The model was trained using the SEN2NAIP dataset and evaluated in Cuyama Valley, Santa Barbara County, California, USA, using NAIP imagery, and through zero-shot deployment in Kenya using PlanetScope as a reference. The generated 2.5 m products represent spatially enhanced estimates rather than direct observations of 2.5 m information. Performance was assessed using mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and edge preservation index (EPI), with bicubic interpolation and OpenSR, a latent-space diffusion model, as baselines. In Cuyama Valley, SR3 improved SSIM from 0.246 to 0.389 and PSNR from 8.2 to 14.0 dB while reducing MSE from 0.1515 to 0.0304 compared with bicubic interpolation. SR3 also outperformed OpenSR, achieving SSIM of 0.384 and PSNR of 14.0 dB compared with 0.249 and 8.19 dB, respectively. In the zero-shot Kenyan evaluation, bicubic interpolation produced lower MSE, higher PSNR, and higher SSIM than both diffusion models at Kieni Forest and Dedan Kimathi University of Technology, while SR3 demonstrated competitive edge preservation. These findings show that pixel-space diffusion can enhance Sentinel-2 spatial detail within the training domain, while zero-shot performance remains sensitive to cross-sensor and geographic differences. The imagery has potential for precision agriculture, vegetation monitoring, environmental assessment, land-cover mapping, and coastal and marine science, where enhanced spatial detail can complement Sentinel-2’s temporal coverage.

Keywords: satellite image, super-resolution, diffusion models, Sentinel-2 SR3, remote sensing, multispectral imagery, spatial resolution enhancement, earth observation

DOI https://doi.org/10.33803/yvucyrPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/yvucyr ↗

56.Spatio-Temporal Analysis of Surface and Groundwater Interactions in the Amboni River Basin, Kenya, Using SWAT-MODFLOWAbstract

Johanna Anyesi Wanjala^1*⁠iD, Bartholomew Thiong’o Kuria¹, Arthur W Sichangi¹, Charles Ndegwa Mundia¹, Godfrey Ouma Makokha², Andreas Rienow³

Understanding surface water (SW) - groundwater (GW) interactions is essential for sustainable water management, as these exchanges regulate streamflow, recharge, and ecosystem health. However, limited knowledge of their occurrence, distribution, and flow dynamics, exacerbated by climate variability and human activities, has contributed to water insecurity, emphasising the urgent need to investigate SW–GW interactions for reliable freshwater availability. This study addresses the lack of integrated modelling of SW–GW interactions by simulating their dynamics (1993–2022) in Kenya’s Amboni Basin using the coupled Soil and Water Assessment Tool - Modular Groundwater Flow Model (SWAT-MODFLOW). The results were calibrated in three steps: SWAT model with SUFI using one-gauge discharge; MODFLOW model with Parameter ESTimation (PEST) using hydraulic conductivity; and final manual adjustments of key parameters for the coupled model. Using topographic, soil, land-cover, and meteorological data, the simulation showed strong SW–GW interaction, with average groundwater recharge of 3.3 m³/day and discharge of 220.2 m³/day to streams. The spatiotemporal distribution of GW recharge from stream seepage and precipitation showed substantial variation, with most recharge occurring in April–May and October–November. Areas at higher elevations received more rainfall and experienced greater recharge than low-lying regions. GW and SW interaction increased with increasing rainfall. However, it was noted that the recharge to GW has been decreasing since 2016, from a monthly average of 37.46 mm to 19.58 mm, despite variations in rainfall. Effective water resource management in the Amboni Basin requires stricter regulation of groundwater and surface water extraction and conservation efforts to protect water towers and sustain recharge processes.

Keywords: surface water–groundwater interaction, discharge, recharge, coupled SWAT-MODFLOW, Amboni basin

DOI https://doi.org/10.33803/bjhz3rPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/bjhz3r ↗

57.Spatio-Temporal Modeling of Nomadic Pastoralist Mobility Responses to Resource Availability in Northern Somalia Using an Agent-Based ApproachAbstract

Umikaltuma Ibrahim^1*, Mark Kipkurwa Boitt¹, Duncan Maina Kimwatu¹

Mobility is a crucial strategy that enables pastoralists to maximise variable water and pasture conditions. Understanding mobility patterns aids decision-makers to develop sustainable programs to suit pastoralist lifestyles and ensure they are not left behind. Yet, there is little quantitative insight into long-term modeling of mobility patterns in order to identify any shifts across contrasting environmental conditions. To fill this gap, this study developed a calibrated spatial-based agent-based model (ABM) that simulates monthly mobility of pastoral agents over six years (2019-2024) in Tug Der/Nugaal drainage basin in Northern Somalia in response to pasture (Soil-Adjusted Vegetation Index) and water (rainfall and water sources) availability. The resulting monthly agent locations were analyzed to identify emergent spatio-temporal mobility patterns. Model calibration yielded a composite fitness score of 0.630. Peak basin exits occurred in October and April, corresponding to the onset of the rainy seasons. Additionally, the standard deviation ellipse of agent locations expanded substantially, increasing by 42% from 244,779 km² in 2019 to a maximum of 352,449 km² in 2023 before contracting marginally to 347,558.31 km² in 2024. The centroid of the distribution shifted consistently westward over the study period. Direction of annual centroid displacement was calculated as bearing from true north, changing from north-westward (302° 38' 23" - 308° 52' 59") in 2020-2021 to west-southwestward (237° 50' 53") by 2024. Additionally, the rotation angle of the directional structure (also measured from north), increased steadily from 103° 44′ 18″ to 111° 01′ 46″, while agents frequently returned to the basin. These findings demonstrate that environmental variability shapes mobility dynamics in dryland pastoral systems, and reinforce the need to safeguard access to extensive rangelands and movement corridors that enable adaptive responses to shifting resource availability.

Keywords: pastoral mobility, agent-based modeling, calibration, resource variability, spatial analysis, Somalia

DOI https://doi.org/10.33803/jmtpqzPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/jmtpqz ↗

58.A Life Cycle Optimisation Framework for Selecting Maintenance Service Contracts for Cancer Treatment Machines in Resource Constrained Health SystemsAbstract

Stephen Vala^1*, James Wakiru¹⁠iD, Peter Muchiri¹

Cancer treatment machines (CTMs) are central to delivering radiotherapy to nearly half of all cancer patients. In low- and middle-income countries (LMICs), where cancer incidence is rising, governments have invested heavily in procuring these systems without budgeting for future operating costs because of budget constraints. Ensuring uninterrupted machine availability is essential for better clinical outcomes. However, CTMs are complex, capital-intensive assets that cost an average of $4M, require specialised maintenance support, typically provided through Original Equipment Manufacturer service contracts. The World Health Organisation issued guidelines for LMICs to sign these contracts at an average of 10–15% of the acquisition cost annually, which is high. The operational and maintenance funds often cover comprehensive maintenance contracts that include preventive and corrective maintenance throughout the equipment life cycle, even in the early years when failure rates are low. Institutions often use this practice to avoid the stochastic ordering of these services, which requires requisitions every time a service is required, needed due to constrained resources. This leads to inefficient resource allocation and reduces the funds available to sustain long-term equipment performance since these contracts are not based on the life cycle age or operating context This study develops a life cycle-based framework, using simulation modelling, to guide the selection of optimal service contracts for CTMs in resource-constrained settings. The model integrates equipment age, subsystem criticality, operating context and contract characteristics to support evidence-based decision-making. The model is validated using empirical data from a case study carried out at a referral hospital in Kenya. This research is the first to operationalise a life-cycle perspective for CTM maintenance contracting, shifting decision-making from blanket contract adoption to context-specific, age-responsive optimisation. By aligning contract choice with life-cycle characteristics, the framework enables upfront cost savings and strengthens long-term system availability, ultimately improving continuity of cancer care in LMICs.

Keywords: cancer treatment machines, low- and middle-income countries, maintenance service contracts, original equipment manufacturer, world health organization

DOI https://doi.org/10.33803/qc9fcqPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/qc9fcq ↗

59.Comparison of Bacterial Cellulose and Plant-Based Cellulose from Sugarcane Bagasse for Wound Healing ApplicationsAbstract

Wincate Mukami Ngari^1*, Esther Magiri¹, Paul Tanui¹, Samuel Mirie¹

Bacterial cellulose (BC) is widely regarded as an excellent biomaterial owing to its near-100% purity, high crystallinity (often 80-90%), nanofibrillar 3D network, exceptional water-holding capacity (up to 100-200 times its dry weight), mechanical strength, and biocompatibility. However bacterial cellulose (BC) has emerged as a promising biomaterial due to its unique structural properties, a consistent finding across reviews show that BC's industrial scale-up is constrained by the limited number of high-yielding bacterial strains, long fermentation cycles, and costly defined culture media, which alone can account for roughly half to two-thirds of total production expenses. Plant-derived cellulose obtained from sugarcane bagasse has been investigated as a scalable, lower-cost alternative once lignin and hemicellulose are removed through mercerization, bleaching, or alkaline or steam-explosion pretreatment. A study conducted on an engineered guanidine-modified bagasse cellulose hydrogel reached swelling ratios of up to 2,000% and a compressive strength of 31.39 kPa, produced greater than 99.99% growth inhibition of Escherichia coli without leaching of the antimicrobial agent, and maintained fibroblast (NIH3T3) viability above 76%. In another separate study, a fabricated bagasse-derived cellulose membrane absorbed water rapidly (roughly 600% swelling within 60 minutes), supported human dermal fibroblast proliferation, showed no in vitro cytotoxicity, and caused no notable skin irritation in an in vivo rabbit model. Collectively, these findings support the view that properly processed sugarcane bagasse cellulose hydrogels and membranes can match several of the functional benchmarks associated with bacterial cellulose while offering decisive advantages in raw-material cost, feedstock abundance, and production scalability, since bagasse is a widely available agricultural residue rather than a fermentation product constrained by strain availability and medium cost. This review compares plant-based cellulose from sugarcane bagasse and positions appropriately purified bagasse-derived cellulose, as a viable sustainable alternative to bacterial cellulose for clinical wound-care applications.

Keywords: plant cellulose, bagasse, wound-healing, bacterial cellulose, hydrogel

DOI https://doi.org/10.33803/3kyvndPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/3kyvnd ↗

60.Design and Development of a Low-Cost Multi-Channel EMG/EOG Signal Acquisition Module for Human–Machine InteractionAbstract

Moses Mweu Mulinge¹, Brian Bundi Kithinji¹⁠iD, Joseph Muguro^1*⁠iD, Brian Ndambia¹

This project addresses the challenges faced by individuals with motor impairments who may experience difficulties interacting with conventional human–machine interfaces that rely on physical input devices such as keyboards, mice, or touch interfaces. Biosignal-based human–machine interaction provides an alternative approach by utilizing physiological signals such as Electromyography (EMG) and Electrooculography (EOG) to translate muscle activity and eye movements into control commands. However, existing biosignal acquisition systems may rely on specialized or relatively expensive hardware, while some are limited to a single biosignal modality, reducing their accessibility and flexibility for assistive applications. To address this gap, this project presents the design and development of a low-cost, multi-channel EMG and EOG biosignal acquisition module for assistive human–machine interaction. The system is designed to acquire electrical signals generated by muscle activity and eye movements and convert them into digital signals for subsequent processing and classification. The acquisition front-end utilizes the AD620 instrumentation amplifier for differential amplification, while the LM324 operational amplifier is employed for signal conditioning and filtering to reduce noise and unwanted frequency components. The conditioned signals are digitized using the ADC of the Seeed Studio XIAO ESP32-C3, which enables wireless transmission through Wi-Fi and Bluetooth to an external processing platform. The developed module integrates signal amplification, filtering, analog-to-digital conversion, and wireless transmission into a compact and affordable platform. The system is intended to provide a flexible foundation for EMG and EOG signal processing and classification, with potential applications in assistive human–machine interaction, cursor control, virtual keyboards, and other accessibility-oriented interfaces.

Keywords: EMG, EOG, ADC, XIAO ESP32-C3

DOI https://doi.org/10.33803/andk5kPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/andk5k ↗

61.Design and Implementation of an Automatic Telemetry-Enabled Medication Dispensing UnitPoster

Dorothy Gicheru^1*, Kenneth Kimosop^1*, Salmon Musina^1*, Ivonne Mayaka^1*

Background. Patients on several medicines often miss doses or take the wrong pill. Most automatic pill dispensers run on a timer and never check what was loaded or whether a pill actually fell. We set out to build a low-cost dispenser that checks each refill, confirms every dose and alerts both patient and caregiver.

Methods. We designed and built the dispenser in a university laboratory in Kenya around an ESP32 microcontroller, with two 3D-printed PETG carousels turned by stepper motors. An RFID reader checks each medicine container before a refill, a real-time clock keeps the schedule without internet, and a piezoelectric sensor in the chute detects each falling pill. The ESP32 keeps a pill count that falls by one with each confirmed drop. Each event is sent over Wi-Fi to a web dashboard, and a Bluetooth wristband vibrates to alert the patient.

Results. We have a working prototype. Both carousels turn on schedule, and each releases one pill when its slot exit lines up with the chute. A refill with an unregistered RFID tag stays locked, and each dispensing event reaches the web dashboard while the wristband vibrates. In a first bench trial of 10 scheduled doses, 7 were dispensed correctly (70 percent). Testing over 50 scheduled doses and 10 wrong-tag refills is in progress, as we work to raise this rate towards our targets of a drop within 5 seconds of its set time and at least 95 percent of drops detected.

Conclusions. The prototype shows that a low-cost dispenser can check each refill and confirm each dose, improving on timer-only dispensers. Raising its dispensing reliability is the next step towards use by patients at home and their caregivers.

Keywords: automated medication dispensing, RFID, piezoelectric sensing, ESP32, telemetry

62.Energetics of the Spontaneous Beating of Motor-Driven FilamentsAbstract

Jane Wanja Karanja¹, Douglas Kagoiya Ng’ang’a¹, Takahiro Nitta^1*⁠iD

Cilia and flagella beat through the collective action of molecular motors, driving fluid transport and cell motility, and are an attractive template for engineered actuators. The follower force model captures spontaneous beating in minimal reconstituted systems, but treats motor activity as a uniform, constant force density and so cannot resolve how individual motors supply and dissipate the energy that sustains oscillation. We simulated a microtubule (MT, 5–30 μm long) clamped at its minus end and driven by randomly distributed surface-immobilized kinesins (5–50 motors/μm²). Each motor was modelled as a linear spring (k = 100 pN/ μm) with load-dependent velocity (unloaded 0.8 μm/s, stall force 5 pN) and force-dependent detachment (detachment force 7 pN). Dynamics were obtained by solving overdamped equations, with hydrodynamics included only as local isotropic drag. Beating emerged from a localized buckling instability once the motor force exceeded a critical threshold set by MT length and motor density, with frequencies of 30–50 mHz. Bending waves travelled from the clamped end to the free end at ~0.8 μm/s, the unloaded kinesin velocity. Frequency scaled with motor density with an exponent of 0.27 ± 0.01, much weaker than the 4/3 predicted by the follower force model; we attribute this discrepancy to the mechanical interaction between the filament and kinesin. Amplitude scaled with an exponent of −0.27 ± 0.02, consistent with the scaling of the buckling wavelength. These results show that motor protein mechanics govern the energetics of wave-like beating in actively driven filaments and can guide the design of micro- and nanoscale devices for fluid pumping and mixing.

Keywords: cilia, microtubule, kinesin

DOI https://doi.org/10.33803/ee5s33Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ee5s33 ↗

63.Machine Learning-Based Risk Stratification of Depression, Anxiety, and Suicidal Ideation Among Adolescent Girls and Young Women in KenyaAbstract

Jacqueline Akelo Gogo^1*⁠iD, Zakayo Morris Ndiku¹⁠iD, David Muchangi Mugo²⁠iD, Dominic Kitavi Makaa¹⁠iD

Mental health conditions remain insufficiently examined in sub-Saharan Africa, particularly among adolescent girls and young women who are out of school. Existing screening tools often rely on narrow predictors or single outcomes, leaving important gaps. To address these limitations, machine learning methods were applied to identify concurrent risk of depression, anxiety, and suicidal ideation in Kiambu County, Kenya. A community-based cross-sectional survey recruited 13,396 participants aged 15 to 29 years. The prevalence of mental health conditions was substantial: 39.9% met criteria for depression, 20.2% for anxiety, and 42.3% reported suicidal ideation. Out-of-school participants consistently showed greater vulnerability. Depression was assessed using the Patient Health Questionnaire–9 with a threshold score of ten or higher, anxiety with the Generalized Anxiety Disorder–7 using the same threshold, and suicidal ideation with the Columbia Suicide Severity Rating Scale screener referencing the past month. Random Forest, XGBoosting, Least Absolute Shrinkage and Selection Operator regression, and Support Vector Machine were compared with a Stacked Ensemble-Adaptive model that combined these learners using logistic regression as a meta-learner, weighted by validation performance. Performance was evaluated through nested tenfold cross-validation with a held-out validation set. The ensemble achieved area under the receiver operating characteristic curve values of 0.89 for depression, 0.84 for anxiety, and 0.83 for suicidal ideation, modestly outperforming the strongest single comparator, XGBoosting, which achieved 0.86 for depression. Calibration slopes between 0.95 and 0.98 indicated strong agreement between predicted and observed risks, and decision curve analysis showed positive net benefit across thresholds. Variable importance analyses highlighted adverse childhood experiences, loneliness, and maladaptive coping strategies as key risk factors, while self-esteem and family support were protective. A three-tier framework was defined: low risk (<0.33), moderate risk (0.33–0.66), and high risk (>0.66). External validation in other settings is required before broader generalisation since we confined our recruitment to Kiambu County.

Keywords: adolescent girls and young women, depression, anxiety, suicidal ideation, machine learning, Kenya

DOI https://doi.org/10.33803/n2pfbsPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/n2pfbs ↗

64.Monitoring and Evaluation Practices and the Performance of Maternal Health and Family Planning Projects: Evidence from the NPI Expand Project in Narok County, KenyaAbstract

Zeytun Adan^1*, Nehemiah Kiplagat¹, James Mushori¹

Monitoring and evaluation remains a central pillar of project management practice, yet its contribution to maternal health and family planning outcomes in pastoralist and culturally conservative settings has received limited empirical attention. This study examined the influence of monitoring and evaluation on the performance of maternal health and family planning projects under the New Partnerships Initiative Expand Project in Narok County, Kenya, and tested whether cultural attitudes moderated this relationship. Anchored on the Results Based Management Framework and Hofstede's Cultural Dimensions Theory, the study adopted a mixed methods approach combining a cross sectional descriptive survey with a correlational design. A sample of 292 respondents drawn from Community Health Volunteers, Health Care Providers, and the project team responded to structured questionnaires, complemented by Key Informant Interviews and Focus Group Discussions. Monitoring and evaluation recorded a composite mean of 3.76, indicating implementation to a large extent, with use of monitoring results as the strongest dimension and planning and design as the weakest. Cultural attitudes recorded a composite mean of 4.06, with gender norms in health decisions emerging as the strongest dimension. Correlation analysis revealed a very strong positive relationship between monitoring and evaluation and project performance (r = 0.897, p < 0.05), while simple linear regression established that monitoring and evaluation explained 80.5 percent of the variation in performance (R² = 0.805, F = 1194.176, p < 0.05). Moderation analysis using Hayes' PROCESS Macro Model 1 revealed that although cultural attitudes had a significant direct effect on performance (β = 0.180, p = 0.023), the interaction between monitoring and evaluation and cultural attitudes was not statistically significant (β = -0.052, p = 0.158).Monitoring and evaluation is the strongest predictor of project performance, remaining effective across cultural contexts. The study recommends institutionalizing monitoring, strengthening feedback loops, and pairing systems with culturally responsive engagement.

Keywords: monitoring and evaluation, cultural attitudes, project performance, maternal health, family planning, Narok county

DOI https://doi.org/10.33803/54cvp2Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/54cvp2 ↗

65.Phytocompounds of Cajanus Cajan as Potential Anti-Breast Cancer Agents: an Analysis Through Network Pharmacology and Molecular DockingAbstract

Douglas Kahura Njuguna^1*, Benson Ongarora¹, Paul Tanui¹, Paul Sang¹

Breast cancer is a leading cause of death among women. Cajanus cajan has been shown to have anticancer activity on breast cancer cells. This study explored the anti-breast cancer potential of Cajanus cajan through network pharmacology and molecular docking techniques. Phytocompounds of Cajanus cajan were retrieved from literature and the IMPPAT 2.0 database. The compounds were filtered through Lipinski's rule and ADMET properties, and their target genes were predicted using Swiss Target Prediction and the similarity ensemble approach (SEA). Differentially expressed genes in breast cancer were retrieved from the Gene Expression Omnibus, GeneCards, and OMIM databases. Protein-Protein-Interaction (PPI) network was constructed using the String database, and analysis was done in Cytoscape software. The DAVID database was used for Gene Ontology enrichment and KEGG Pathways analysis. Molecular docking was performed using PyRx software and visualized in Biovia Discovery Studio. Ten compounds were identified through filtering, which include; 5-Hydroxy-7-methoxy-2-(4-methoxyphenyl)-2,3-dihydrochromen-4-one, (2s)-5-Hydroxy-7-methoxyflavanone, Apigenin, luteolin, Naringenine, Quercetin, Formononetin, cajanol, Cajanin from leaves and five from the roots namely; cajanol, Cajanin, 2'-Hydroxygenistein, Daidzein and Genistein. PPI network analysis revealed nine core target genes, including SRC, PIK3CA, EGFR, PIK3R1, IGF1R, ESR1, CTNNB1, PTK2, and AKT1. These genes were found to be involved in various biological processes such as positive regulation of phosphatidylinositol 3-kinase, negative regulation of apoptotic processes, and positive regulation of cell proliferation, and they involved endocrine resistance, PI3K-Akt signaling, and EGFR tyrosine kinase inhibitor resistance pathways. In conclusion, Cajanus cajan demonstrates anti-breast cancer potential through multi-target genes and multi-target pathway mechanisms. These findings provide a predictive foundation for further experimental validation of its therapeutic mechanisms.

Keywords: Cajanus cajan, network pharmacology, molecular docking, gene ontology, KEGG pathway analysis

DOI https://doi.org/10.33803/4k6fgdPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/4k6fgd ↗

66.A Layered Security Framework and Metrics for Analysing Cross-site Scripting AttacksFull paper

Nevy Kimani Maina^1*⁠iD, Geoffrey Muchiri Muketha^1*, Jane Wanjiru Njuki^1*

Single Page Applications (SPAs) have become a dominant architecture for modern client-side web systems by shifting state management, routing, and user interface processing to the browser. Although frameworks such as Angular provide built-in security mechanisms, including contextual escaping, template sanitization, and secure data binding, developers may introduce security weaknesses through practices such as unsafe DOM manipulation, dynamic script loading, and improper use of bypass mechanisms. Existing approaches to SPA security assessment primarily focus on identifying known vulnerability patterns through static analysis, providing limited insight into architectural security weaknesses and accumulated security complexity. This study contributes a Metrics-Driven Layered Security Framework (LSF) for assessing and improving the security architecture of Angular SPAs against Cross-Site Scripting (XSS) attacks. The proposed framework organizes SPA security controls into three architectural layers: the Rendering Layer, Control Logic Layer, and Outer Shell Layer. It introduces seven security metrics, including Sanitization Bypass Density, Unmanaged Active Sink Frequency, Maximum Nesting Depth, Cyclomatic Branching Factor, CSP Policy Strictness Score, Boundary Validation Density, and Dependency Vulnerability Coefficient, which are combined into composite layer risk scores for comprehensive security assessment. The framework was empirically evaluated through controlled experiments involving Angular developers, comparing developers using the LSF-supported Layered Security Metrics Tool (LSMT) against developers performing unaided manual security reviews. The findings demonstrated that the proposed metrics and framework improve developers’ ability to identify structurally complex XSS-related weaknesses across Angular SPA security layers. The results demonstrate that the LSF provides a measurable approach for proactive vulnerability identification, architectural security auditing, and integration into Continuous Integration/Continuous Deployment (CI/CD) security processes.

Keywords: angular, security framework, single page application, cross-site scripting, Sanitization

DOI https://doi.org/10.33803/j5g4agPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/j5g4ag ↗

67.From Knowledge to ImpactAbstract

James Wakiru^1*⁠iD

Introduction. Higher education institutions face growing expectations to address graduate unemployment, technological change, evolving industry needs, and the transition toward Competency-Based Education (CBE). This paper examines how embedding innovation, entrepreneurship, and technology commercialization as an institutional philosophy can reposition higher education institutions as drivers of employment creation, industrial development, and socioeconomic transformation.

Methods. This paper adopts a theoretical and practice-based review approach, synthesizing emerging practices in innovation and entrepreneurship, CBE, university–industry collaboration, and technology commercialization. These dimensions are integrated into an Innovation, Entrepreneurship and Commercialization framework, with Dedan Kimathi University of Technology (DeKUT), Kenya, used as an institutional example to illustrate how the framework can be translated into practice.

Results. The analysis shows the need to shift from a predominantly knowledge- and qualification-centric model to an integrated continuum linking CBE and experiential learning to industry problem identification, applied research and innovation, intellectual property development, incubation and enterprise creation, and technology commercialization and impact. Using DeKUT as an illustrative context, practical implementation requires embedding innovation competencies within curricula, establishing industry-defined student and research challenges, strengthening prototyping and incubation infrastructure, developing technology-transfer and intellectual-property capabilities, incentivizing faculty and student innovation, and creating partnerships for financing, market validation, licensing, and start-up growth.

Conclusion. Embedding innovation, entrepreneurship, and commercialization across teaching, research, and industry engagement provides a practical pathway for transforming higher education institutions from predominantly producers of graduates and research outputs into creators of solutions, technologies, enterprises, and jobs, thereby strengthening their contribution to industrial competitiveness and sustainable socioeconomic development.

Keywords: higher education, innovation, entrepreneurship, technology commercialization, competency-based education, University–industry linkages, job creation

DOI https://doi.org/10.33803/wr7kymPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/wr7kym ↗

68.Loan Default Management and Sustainability of Public–private Funded Micro and Small Enterprise (MSE) Credit Programmes in Nairobi County Government, KenyaAbstract

Makumbi Johnstone Kakuti^1*

Micro and Small Enterprises (MSEs) are central to employment creation, income generation and local economic development in Kenya, yet limited access to sustainable finance remains a persistent constraint to their growth. Public–private funded credit programmes have emerged as an important mechanism for expanding affordable financing to MSEs. Although previous studies have examined credit access, repayment behaviour and loan default among MSEs, limited attention has been given to how loan default management within public–private funded county credit programmes influence programme sustainability. Anchored on the Pecking Order Theory, this study examined the effect of loan default management on the sustainability of public–private funded MSE credit programmes in Nairobi County Government. The study adopted a descriptive correlational cross-sectional design and collected data from county finance and business-development officials and representatives of beneficiary groups using structured questionnaires and interviews. The findings indicated moderate application of loan default management practices. Loan default management was found to have a strong, positive and statistically significant relationship with the sustainability of MSE credit programmes (r = 0.782, p < 0.001). The findings suggest that effective management of loan defaults is an important mechanism for protecting revolving credit resources, maintaining repayment capacity and supporting continuity of MSE financing. The study concludes that strengthening loan default management can contribute to the sustainability of public–private MSE credit programmes and provide continuous financial support to small enterprises. The study recommends that county governments and private financial partners establish coordinated default-management frameworks incorporating early-warning mechanisms, borrower counselling, timely follow-up, dedicated recovery structures and structured loan-restructuring procedures.

Keywords: Rds: loan default management, sustainability of credit programmes, pecking ord

DOI https://doi.org/10.33803/gh38fpPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/gh38fp ↗

69.Mentorship and Professional Identity in Kenya's Tourism and Hospitality EducationAbstract

Rose Burugu^1*⁠iD

Higher education is changing rapidly because of technology, AI, globalisation, evolving work patterns, demographics, and career uncertainty. Employers want creative, adaptable problem-solvers committed to lifelong learning, while training institutions aim to equip students with a broad skill set. A key challenge is the industry's unmet demand for soft skills and the difficulty of integrating students into the service sector because of gaps in mentorship training and industry alignment. This raises questions about whether mentorship methods are sufficient for diverse students and a changing industry. The article discusses how mentorship fosters innovation, independence, and adaptability, preparing students for careers in hospitality and tourism. Despite the industry requiring quick learners with strong communication, networking, customer service, sales, and opportunity-recognition skills, graduate unemployment and limited entrepreneurship show that academic training alone isn't enough, making mentorship vital as programmes focus on practical, competency-based skills. A survey of 166 first-year students (2025-2026) examined expectations regarding mentorship duration, topics, mode, support, and areas for improvement. Drawing on the literature and initial descriptive data, this study proposes a mentorship framework integrating experiential learning, industry engagement, creativity, and autonomy. Results indicate that mentorship boosts personal growth, confidence, creativity, independence, decision-making, and responsibility. A structured program is recommended, with mentors as facilitators who challenge, inspire, connect, give feedback, expand networks, and delegate tasks. Suggestions include embedding student-driven mentorship programs into curricula, improving mentor training, strengthening industry ties, promoting interdisciplinarity, and using digital platforms to broaden mentoring. Aligning mentorship with experiential learning from Year One helps students develop professional identities, explore careers, and apply knowledge. As students and workplaces change, mentorship should transition from random, imposed guidance to intentional support that promotes independence, innovation, and adaptability, considering each student's diversity. This method helps graduates generate opportunities, mentor others, and drive continuous development.

Keywords: mentorship, mentors, mentorship methods, professional identity, soft skills

DOI https://doi.org/10.33803/sjy4tbPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/sjy4tb ↗

70.Observability Limits of Downhole State Estimation from Surface Measurements in Geothermal DrillingAbstract

John Petu^1*⁠iD, Charles Kagiri¹⁠iD, Samuel Macharia Kang'iri^1*⁠iD

Stick slip and whirl wreck bits and bottom hole assemblies, and in geothermal wells heat limits the downhole electronics that could watch them. Every surface based monitoring scheme therefore presumes the downhole state can be recovered from surface signals, and when it cannot, the mitigation acts on a guess. Observers, filters and learned estimators built on that presumption are numerous, but the conditions for recovery have never been established, and their models are calibrated for oil and gas wells, not geothermal rock and temperature. This work makes recoverability the object of study, and asks it of individual physical quantities rather than the state as a whole. A coupled axial, torsional and lateral drillstring model with a temperature dependent bit to rock law is probed by perturbing each quantity and measuring how strongly surface torque, top drive speed and hook load respond. Because stick slip makes the motion non smooth and whirl makes it chaotic, it is taken over short windows, averaged over many initial conditions and referenced to sensor noise, separating what is observable in principle from what is recoverable in practice. Unscented Kalman filter error confirms the measure predicts achievable accuracy; the model reproduces published laboratory and field vibration records. The analysis partitions the weight on bit and rotary speed plane into identifiable, weakly identifiable and unrecoverable regions, and reports where sampling rate and latency lose identifiability. The partition is uncomfortable: bit rock interface quantities are recoverable mainly inside the stick slip region operators work to eliminate, while formation strength is recoverable in any regime given 15 s. Raising bottom hole temperature moves the boundaries, so an oil and gas partition does not transfer. This tells a tool designer which quantities justify a downhole sensor, which a surface estimator can supply, and at what sampling rate and latency.

Keywords: drillstring vibration, observability analysis, geothermal drilling, downhole state estimation

DOI https://doi.org/10.33803/dgzvrsPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/dgzvrs ↗

71.Post-Fire Strength Recovery and Durability of High-Strength Concrete Reinforced with Bamboo Leaf Ash and Polypropylene Fibers - Phase IPoster

Vincent Otieno Oriyo^1*⁠iD, Erastus Karoki Kabando¹⁠iD, Kiplagat Chelelgo¹⁠iD

Fire-induced deterioration remains a major threat to the safety of high-strength concrete (HSC) structures. While HSC exhibits superior mechanical properties, its dense microstructure makes it susceptible to explosive spalling and residual strength loss under elevated temperatures. These challenges compromise structural integrity and occupant safety after fires. The CTIF World Fire Statistics Report highlights the continuing toll of structural fires, causing loss of life, property damage, and economic disruption, underscoring the need for more fire-resilient infrastructure. This study therefore investigates the potential of bamboo leaf ash (BLA) and polypropylene (PP) fibres to enhance the post-fire strength recovery and durability of HSC. In this first phase of the study, we aim to determine the optimum PP fibre dosage for 55 MPa HSC that can effectively mitigate explosive spalling. This phase provides a basis for subsequent investigations into the synergistic effects of BLA and PPF in post-fire strength recovery of HSC. Therefore, 100mm × 100mm × 100mm concrete cubes were prepared with PP fibre dosages of 0%, 0.15%, 0.20%, 0.25% and 0.30% by concrete volume. The specimens were air-dried for 14 days, then heated to 600 °C by simulating the ISO 834 standard fire curve. The initial moisture content was 1.8–2.7%, below the 3% threshold specified in Eurocode 2 beyond which explosive spalling becomes likely. No explosive spalling was observed in the specimens with PP fibre. Consequently, a U-Net semantic segmentation model was used to analyse high-resolution images and determine crack density and crack area ratio. The Derringer–Suich desirability function was used to determine optimum PP fibre dosage against five responses: crack density, crack area ratio, mass loss, initial & residual compressive strength. The optimum PP fibre dosage was determined to be 0.26%. This study establishes that deep-learning models can be used to quantitatively assess fire-induced cracking in mild concrete spalling.

Keywords: high-strength concrete, bamboo leaf ash, polypropylene fibers, fire resistance, post-fire recovery, explosive spalling, U-NET semantic segmentation model

DOI https://doi.org/10.33803/d2c4rpPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/d2c4rp ↗

72.The Limits of Data-Sheet Calibration for Micro Turbojet Performance Prediction at AltitudeFull paper

Maxmillian Mwenda Mugambi^1*⁠iD, Paul Waweru Njeri^2*, Titus Mulembo¹, Josphat Tanui³

Micro turbojet performance models are commonly calibrated against manufacturer data sheets and subsequently used to predict engine behaviour across flight conditions for which experimental data are unavailable. This study investigates whether such data-sheet calibration uniquely determines the underlying engine model and quantifies the consequences of parameter non-identifiability for off-design performance prediction. The AMT Olympus HP is modelled using a single-spool thermodynamic cycle incorporating temperature-dependent gas properties, scaled component maps, turbine-to-compressor heat transfer, and both choked and unchoked nozzle operation over Mach numbers from 0 to 0.8 and altitudes up to 8000 m. Calibration against the available manufacturer data admits parameter combinations with compressor efficiencies ranging from 0.66 to 0.79. When internal heat-transfer conductance is allowed to scale with air mass flow to the power 0.8, the predictions at 8000 m static have a thrust spread of 4.5% and an installed thrust-specific fuel consumption (TSFC) spread of 9.7%. The altitude trend in the internal parasitic heat fraction is described by a flow-corrected dimensionless group, $\Lambda_{0.8} = \Lambda\left(\frac{\dot{m}_a}{\dot{m}_{a,\mathrm{ref}}}\right)^{0.8}$. In a within-model comparison, this group tracks the representative model's nonlinear heat-fraction trend to within 3.3% over the stated maximum-rpm envelope; this is not independent experimental validation. A hypothetical compressor-exit total-temperature constraint centred on the median of the calibration family reduces the predicted thrust spread at 8000 m static to 1.5%; scanning the possible constraint-window positions gives a range of 0.2% to 1.7%. No independently verified numerical compressor-exit temperature is imposed in this analysis, so this is a sensitivity result rather than an applied experimental constraint.

Keywords: micro turbojet engine, model calibration, parameter identifiability, gas turbine heat transfer, altitude performance prediction

DOI https://doi.org/10.33803/2m3ue6Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/2m3ue6 ↗

73.Automated Measurement of Microtubule Motility Parameters Using a Deep Neural Network Trained with Physics Based Synthetic DataAbstract

KIMANDE Jethro Wanjala¹, Takahiro Nitta^1*⁠iD

Biomolecular motors carry molecular cargo or generate force. For example, kinesin transports intracellular cargos by walking along filaments called microtubules. In in vitro motility assays, microtubules are made to glide over surface-immobilized kinesins. The accurate measurement of microtubule dynamics, such as length, is critical for understanding intracellular transport. In measuring these microtubule parameters, researchers often rely on manual methods which are time consuming and less accurate due to human bias. Advanced algorithmic tracking methods can fit 2D Gaussian models to localize filaments with subpixel resolution but remain sensitive to fluorescence quality and manual intervention. Machine learning models can offer automated, robust, and high-throughput tracking solutions. However, the primary bottleneck in training such models is the limited availability of large annotated experimental data. To bridge this gap, a dataset of 12,000 synthetic images was generated using off-lattice Monte Carlo simulations of microtubules gliding over kinesin motors, with known lengths of 5 um to 11 um, a speed of 850 nm/s, and a diameter of 25 nm. To mimic the observable microtubules in fluorescence microscopy, the generated matrices were subsequently convolved with optical blurring (via Point Spread Functions) and digital camera sensor noise with a mean intensity of 50. These physics-based simulations provided ground truth data, that was used to train a deep U-Net convolutional neural network for automating the segmentation and measurement of the microtubule physical properties. A test on 50 synthetic single microtubule images, our method achieved a 0.069 pixel centerline deviation and 0.04 pixel length loss, versus 0.087 and 0.012 with conventional method. This automated approach is expected to reduce human bias, ensuring reproducible, high-throughput, and sub-pixel measurements, extendable to other structures like actin filaments. In addition, this research is expected to accelerate the automated design and testing of active nanoscale transport systems for motor proteins dependent biosensor technologies.

Keywords: deep learning, nanoscale measurements, synthetic data

DOI https://doi.org/10.33803/n77v28Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/n77v28 ↗

74.Autonomous Navigation and Edge-AI Telemetry Optimization in Planetary UGVs: A Case Study of the Vulcan Mars RoverAbstract

Kiptanui Shalom^1*, Victoria Rotich¹, Brian Fundi¹, John Khaemba¹, Allen Wachio¹, Collins Kiprop¹, Shadrack Kinsimba¹, Joan Ouma¹

The development of Unmanned Ground Vehicles (UGVs) for extreme environments requires robust mechanical stability, localized autonomy, and resilient telemetry. This study presents a 4.8 kg terrestrial analogue prototype engineered for resource-constrained field robotics. Operating within limits of 65 × 50 × 30 cm, the platform employs a rocker-bogie suspension system with differential steering, proven capable of delivering a 1.3 kg payload across rough terrain. It is driven by a closed-loop PID control architecture operating at ~30 Hz, utilizing upgraded motor drivers and isolated dual-battery topologies to prevent thermal failures. For autonomy, the architecture utilizes edge computing, abandoning offsite processing. A custom YOLO11s computer vision model was trained on 5,825 source images across three classes (hammer, tennis ball, traffic cone), expanded threefold through augmentation (horizontal/vertical flips, ±15° rotations, and saturation adjustment) and validated on a held-out set of 1,293 images, achieving a 91.8% mAP50, 90.5% precision, 86.6% recall, and an F1 score of 88.5%. By calculating bounding box widths against known target dimensions, the onboard computer estimates relative distances, enabling the rover to halt at safe thresholds and execute sequential waypoint navigation. The full navigation sequence completed with a 100% success rate in a single trial; multi-run statistical validation is ongoing. To resolve severe network congestion, the communication framework utilized a ROS2 "Best Effort" QoS policy routed through an elevated omnidirectional antenna. This measured reduction in packet-drop snowballing achieved sub-100 ms video telemetry latency and stable +45 dB operational link margins during continuous 15-minute field tests in non-ideal outdoor environments at an 80-meter range. The platform successfully demonstrates that decentralized edge computing and optimized RF topologies enable reliable autonomous operations in resource-constrained environments.

Keywords: Edge-AI, autonomous navigation, unmanned ground vehicles, field robotics, ROS2

DOI https://doi.org/10.33803/j7xuybPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/j7xuyb ↗

75.Bi-Objective Genetic Optimization of Camera Poses for Targeted Surface ReconstructionFull paper

Andrew Kibor Chesang^1*⁠iD, Samuel Macharia Kang'iri^1*⁠iD

Viewpoint selection for Structure from Motion has historically split into two paradigms: Camera Network Design (CND), which optimizes sensor configurations offline against a prior object model by propagating pixel uncertainty into 3D covariance, and Next Best View (NBV) planning, which sequentially selects viewpoints online from an occupancy representation built during acquisition. Though each has begun borrowing from the other, both still aggregate reconstruction criteria across the entire object, limiting scalability for geometrically complex targets.This work bridges the two by localizing sensor placement to depth-map-derived Regions of Interest (ROIs), fusing CND’s prior-driven uncertainty modeling with NBV’s targeted, region-scoped efficiency. For each ROI, a bi-objective NSGA-II formulation jointly maximizes inter-view feature-matching consistency and minimizes triangulation uncertainty via an A-optimality criterion, subject to photogrammetric and geometric constraints. Pareto-efficient configurations, selected via pseudo-weighting, were evaluated through a full photogrammetric reconstruction pipeline of a rendered crashed car model against a uniformly sampled baseline. All three pseudo-weight configurations outperformed the baseline in reconstruction accuracy and point-cloud density using only 22 viewpoints versus 71, demonstrating that combining CND’s covariance-optimal rigor with NBV’s localized efficiency yields more information-dense acquisitions than exhaustive uniform sampling, with applications to forensic documentation, stockpile estimation, and archaeological recording.

Keywords: multi-objective optimization, genetic algorithms, structure from motion, camera network design, next-best-view planning, 3D reconstruction

DOI https://doi.org/10.33803/69b7syPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/69b7sy ↗

76.Digital Twin-Based Optimization and Energy Efficiency Analysis of an Automated Tea Blending ProcessAbstract

Timothy Kabugi Gitahi^1*

Value addition in the agro-industrial sector continues to be a very important economic activity in Kenya among its major export crops. Manual operations in most tea blending plants suffer from inefficiencies, mechanical wear, and material blockages due to the poor performance of open-loop feeding processes. In this paper, a high-fidelity Digital Twin architecture for the automation of a tea blending plant has been developed using Siemens DesignCenter NX and TIA Portal to achieve accurate closed-loop mass balance control and electrical power demand monitoring. The physical process was created through a 3D simulation within the NX Mechatronics Concept Designer physics workspace, modelling a blending chamber receiving materials from two automated material hoppers. An industrial control layer was designed using Structured Control Language (SCL) compiled for a virtual SIMATIC S7-1500 CPU to dynamically adjust dual conveyor speeds based on live sensor tracking. Software-in-the-Loop (SIL) virtual commissioning was executed via S7-PLCSIM Advanced over a local TCP/IP single-adapter topology. Real-time trace logs were parsed in MATLAB to extract transient velocity profiles, calculate electromechanical shaft power curves, and model network behaviour inside DIgSILENT PowerFactory. SIL virtual commissioning results demonstrated that the adaptive SCL logic, under the simulated operating conditions, reduces the observed blockage frequency from 4.2 events per hour to zero, maintains a target blending-ratio error within ±0.5% with a settling time under 2.4 seconds, and reduces peak transient active power demand on the plant grid from 16.9 kW to 14.5 kW, corresponding to a 14.2% reduction relative to the baseline peak demand. This integrated approach provides a low-risk virtual commissioning pathway for local agricultural processing facilities transitioning to Industry 4.0 frameworks.

Keywords: industry 4.0, digital twin, virtual commissioning, process control, energy efficiency, tea blending plant

DOI https://doi.org/10.33803/5w4wq8Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/5w4wq8 ↗

77.Vibration Fault Diagnostics in MotorsPoster

Thoithi Joan Mumbi^1*⁠iD, Kennedy Karuri^1*⁠iD, Dennis Kipruto¹, Abdulrhaman Hassan¹, Eric Mumo¹, Samuel Macharia Kang'iri^1*⁠iD

Background. Electric motors drive critical equipment such as pumps and fans, and their vibration signatures change before mechanical failure. Manual inspection can miss sudden faults, while cloud-based monitoring adds latency, bandwidth cost, and network dependence. This project develops an offline edge-AI classifier on a low-cost STM32 microcontroller for real-time motor fault classification without cloud infrastructure.

Methods. A Baldor Reliance 0.33 HP, 400 V, 50 Hz, four-pole three-phase induction motor (1425 RPM rated) was run at rated speed on a test rig to compare a healthy baseline with induced shaft misalignment and structural looseness. A triaxial ADXL345 accelerometer, rigidly mounted on the motor casing, was sampled at 3200 Hz over SPI by an STM32F411CEU6 on a custom KiCad PCB shield. For each 2048-sample window, the firmware removes the DC offset, applies a Hann window, and computes the CMSIS-DSP FFT magnitude spectrum, keeping 256 bins (up to 400 Hz) per axis. The log-scaled, per-window-normalised spectrum feeds a compact 1-D CNN (three convolutional layers of 8, 16 and 16 filters, global average pooling, three-class softmax; about 1.6k parameters). The network will be trained offline in Keras with whole recording sessions held out for testing, quantised to INT8, and deployed via ST X-CUBE-AI.

Results. Data collection and model development are in progress, so classification results are not yet available. Accuracy for the healthy, misalignment, and looseness conditions, with the model’s memory footprint and inference time on the STM32F411, will be reported at the poster session.

Conclusions. Low-cost microcontrollers can replace cloud-dependent condition monitoring platforms through localized, real-time vibration diagnostics. Combining direct adhesive sensing, DSP feature extraction, and quantized neural networks eliminates network latency and inspection delays, delivering an accessible predictive maintenance tool for industrial operators.

Keywords: condition monitoring, Edge-AI, vibration diagnostics, STM32, machine learning quantization, predictive maintenance

DOI https://doi.org/10.33803/ea8udxPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ea8udx ↗

78.Influence of Stakeholder Engagement on Performance of Public-Private Partnership Road Infrastructure Projects :Case of Nairobi ExpresswayAbstract

Onyinkwa Caleb Mounde Nyambega^1*

Public-Private Partnerships are key to addressing Kenya's infrastructure challenges.However,PPP projects suffer from poor monitoring and evaluation, including financial inefficiencies, poor stakeholder engagement, and weak risk governance. The study investigated the influence of stakeholder involvement on the performance of Public-Private Partnership Road infrastructure projects, with a case of the Nairobi Expressway. The parameters used to gauge Stakeholder Engagement included:Communication Strategy, Participation level,Information sharing and Stakeholder feedback.Stakeholder Theory guided the study. The research design was descriptive correlational, with a population of 50,050 people comprising 50 institutional populations and 50,000 road users. Purposive and simple random sampling were employed to select the final sample of 150 respondents from institutional and road user respondents respectively. Structured questionnaires with Likert scales and semi-structured interviews with 100 road users were used for data collection. Quantitative data were analysed using descriptive and inferential statistics, including Pearson’s correlation and multiple linear regression. The result showed strong positive correlation with project performance (r=0.718) and explained 51.6% of performance variance (β=0.487, p=0.000). The study concluded that stakeholder engagement significantly influences PPP performance. Limited participation, weak communication, and inadequate feedback mechanisms reduced project acceptance and satisfaction. The study recommended that: Amend PPP laws to require mandatory stakeholder consultation throughout project lifecycles. Establish permanent road user forums for ongoing engagement, develop publicly accessible project information portals, and institutionalize community participation in toll pricing reviews to ensure decisions balance financial sustainability with public affordability. The study recommended The National Treasury - Public-Private Partnerships Directorate should amend Kenya's Public-Private Partnership Act (2021) to include meaningful stakeholder consultation as an enforceable and legally binding requirement throughout the entire lifecycle of PPP projects, from planning through implementation to operations. Keywords: Public-Private Partnerships, Nairobi Expressway, and Stakeholder Engagement, Project Performance.

Keywords: public-private partnerships, Nairobi expressway, stakeholder engagement, project performance

DOI https://doi.org/10.33803/g968t8Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/g968t8 ↗

79.Internet of Things Based Temperature Sensors Technology in Monitoring Maturity of ConcretePoster

Jesse Pereso^1*⁠iD, Erastus Kabando Karoki¹⁠iD, Kiplagat Chelelgo¹⁠iD

The concrete maturity method offers a non-destructive alternative for estimating compressive strength development of concrete. This method considers the combined effect of temperature and time on concrete strength gain. However, this method has received limited attention in developing countries. Over the years, temperature monitoring techniques have been done manually making it incredibly tedious, with engineers manually executing multi-step routines around the clock. This study evaluates the application of Internet of Things (IoT) based temperature monitoring, to remotely collect temperature data of concrete in real-time stored in a database. The study employed cement class CEM I 42.5N in concrete, with water to cement ratio of 0.40. Cube specimens of 100mm and 150mm were embedded with DS18B20 digital temperature sensors .The cubes were subjected to isothermal curing regimes at 5oC,20 oC, and 40 oC, over a 28 day period. Compressive strength test was performed at 3, 7, 14, 21, and 28 days. From the compressive strength data, the rate of strength gain k ,limiting strength Su and the age of onset of strength development to, were determined using Carino`s hyperbolic maturity function. Solver analysis was used to maximize the coefficient of correlation value R2 ,while varying values of Su ,k and to ,in order to obtain best fit of the hyperbolic function of strength. The Arrhenius plot was then used to quantify for activation energy using the hyperbolic strength gain constant. The activation energy (Ea) ,for the mixture was then determined as 43.93kJ/mol for equivalent age maturity function. The equivalent age te ,was then determined using the temperature profile derived from the temperature sensors, thereafter calculation of strength maturity curves, based on Arrhenius equivalent age indices. The results showed that, IoT-based temperature logging together with the maturity functions, has potential for supporting and monitoring other construction techniques, helping engineers in decision making.

Keywords: maturity, IoT based temperature monitoring, equivalent age, concrete strength

DOI https://doi.org/10.33803/9tvea5Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/9tvea5 ↗

80.Nairobi’s Tourism Dominance: Connectivity, MICE, Business, Gateway and Urban Attractions, 2018–2025Abstract

Peter Ogalo Ogweyo^1*⁠iD

Nairobi continues to be Kenya’s top place for tourism and travel. Still, it is not always clear what evidence ties its long stay at the top to the main drivers. This paper looks at why Nairobi holds its lead. It covers the period from 2018 to 2025. It also checks several parts that shape the city’s role: links to other countries, business and MICE travel, the gateway role, and the city’s urban tourism sites. The study used a desk review method. It relied on existing sources. These included material from the Kenya National Bureau of Statistics (KNBS). It also used state tourism reports, policy papers, and related academic writing. The results point to a clear pattern. Nairobi’s links to other countries matter a lot. The city’s gateway role also helps it pull in visitors and move them onward into Kenya. In 2025, Jomo Kenyatta International Airport (JKIA) logged about 1.72 million international passenger arrivals. At the same time, Moi International Airport recorded about 235,184 international passenger arrivals. These figures show how Nairobi leads in international air access. The study also reports that Nairobi does well in business travel and MICE tourism. In 2025, Kenya held 998 international conferences. Most of these were hosted in Nairobi. This supports the idea that Nairobi is a key base for this type of travel. On top of that, Nairobi offers many urban tourism attractions. These include wildlife experiences. They also include culture and heritage sites, plus ecotourism chances. Visitors can also find options for food, shopping, and entertainment. This mix helps Nairobi compete as a destination. It also helps bring in different kinds of visitors. At the same time, tourism and economic activity are not spread evenly across the country. This can deepen spatial gaps in tourism growth if linkages

Keywords: Nairobi tourism, destination competitiveness, MICE tourism, gateway tourism, international connectivity, urban tourism

DOI https://doi.org/10.33803/ydme6bPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ydme6b ↗

81.A Digital Phenotyping and Mnemonic AI Framework for Addressing Algorithmic Amplification of Harmful Brainrot Content and Cognitive DeclinePoster

Mutongu Levy Nguti^1*, Swabir Mohamed Bwana², Adan Ali Wako³

Background. Algorithmically curated short-form content can repeatedly expose students to salient, rapidly changing stimuli, potentially reinforcing habitual scrolling and reducing opportunities for sustained, goal-directed attention. Reward-related neural circuitry, including the mesolimbic dopamine pathway, supports motivation and reinforcement learning. Digital phenotyping can identify behavioral patterns, while mnemonic AI redirects learning toward engagement.

Methods. We designed a digital phenotyping architecture integrating opt-in social-media signals with TukiStudy in-app behavioral data. Cross-platform feed regulation and in-app signal capture generate passive and active features, which are fused by a non-diagnostic predictive risk model. Model outputs trigger adaptive content or support interventions, followed by human-in-the-loop feedback and retraining.

Results. Across the pilot period, students exposure to harmful content decreased from 70% to 27%, while delayed learning retention increased from 48% to 79%. Mnemonic strategies achieved 72–91% recall, with visual association highest. Post-intervention quiz performance improved across all concepts, suggesting potential benefits for reducing harmful exposure and supporting learning retention.

Conclusions. The integrated framework demonstrated potential to reduce harmful-content exposure while improving learning retention and mnemonic-based recall. Student feedback supported continuous model calibration, with prediction error declining from 0.31 to 0.08 across ten updates. Together, these findings suggest that adaptive digital phenotyping and mnemonic AI may support healthier, more goal-directed learning behaviors.

Keywords: digital phenotyping, mnemonic AI, algorithmic amplification, cognitive decline, adaptive learning, learning retention, behavioral analytics, student wellbeing

DOI https://doi.org/10.33803/gwj9xnPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/gwj9xn ↗

82.Relationship Between Adequacy of School Infrastructure and Implementation of Competency-Based Education by Senior Schools in Nakuru County, KenyaAbstract

Jackline Atieno Ong'injo^1*⁠iD, Betty Tikoko¹, Evelyn Omondi¹

Competency-Based Education (CBE) was extended to Kenyan senior schools in January 2026, raising concern about whether school infrastructure can support its demands. This study examined the relationship between infrastructure adequacy and CBE implementation in Nakuru County senior schools, testing the null hypothesis of no significant relationship. Anchored on Resource-Based View and Systems Theory, it adopted a correlational, convergent parallel mixed-methods design: strands were collected concurrently, analysed separately, and integrated at interpretation by comparing findings for convergence and divergence. Ninety of 485 senior schools were selected by stratified proportionate random sampling (Nassiuma formula); 361 teachers by the Krejcie and Morgan table with simple random sampling; principals and education officers by census; and one Board of Management chairperson per sub-county cluster. Questionnaires were returned by 282 teachers (78.1%) and 72 principals (80.0%); an observation checklist and document review were completed in 74 schools; and 37 key informants were interviewed. Cronbach’s alpha on main-study data yielded α = .909 (teachers) and α = .893 (principals). In SPSS version 31, frequencies, percentages, and means described infrastructure adequacy; Pearson correlation and simple linear regression on the teachers’ dataset tested the hypothesis; principals’ responses, observation records, and interview transcripts were analysed descriptively and thematically to corroborate the quantitative findings. Infrastructure was partially adequate but uneven: classrooms and science laboratories were rated favourably, while workshops, libraries, sports, and digital facilities showed substantial gaps; functional workshops were observed in 21.6% and ICT laboratories in 24.3% of schools. A strong positive association was established, r(280) = .642, p < .001; infrastructure adequacy statistically accounted for 41.2% of the variance in CBE implementation, F(1, 280) = 196.323, p < .001, and the null hypothesis was rejected. As the design is correlational, findings indicate association and statistical prediction, not causation, and can guide infrastructural investment and policy.

Keywords: school infrastructure, competency-based education, senior schools, CBE implementation, mixed-methods research, Nakuru county, Kenya

DOI https://doi.org/10.33803/cwrk2uPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/cwrk2u ↗

83.Relationship Between Peer Supervision and Professional Learning Among Teachers in Junior Secondary Schools in Bungoma South Subcounty KenyaAbstract

Odhiambo Cheril Awino^1*

Peer supervision and professional learning are recognized globally as essential mechanisms for improving instructional quality, teacher effectiveness, and learner achievement. In Kenya, the implementation of the Competency-Based Curriculum in Junior Secondary Schools has created significant demands for continuous teacher development, particularly in pedagogical adaptation, subject specialization, and assessment practices. However, little empirical evidence exists regarding the extent to which peer supervision practices influence teachers' professional learning in the context of Junior Secondary Schools. The study was anchored on the Clinical Theory of Supervision as the anchor theory and the Collaborative Theory of Supervision as the supportive theory. A correlational research design was adopted, employing a quantitative approach to examine the nature of relationships between variables. The target population comprised 90 Junior Secondary Schools, 90 head teachers, and 900 teachers in Bungoma South Sub-County. A multistage sampling technique was applied. Using Krejcie and Morgan's (1970) sample size determination table, a sample of 73 schools was selected from an accessible population of 90 schools through simple random sampling. Subsequently, census sampling was used to include all 73 head teachers from the sampled schools and 146 teachers (two per school), representing 81% of the target teacher population. Data were collected using structured questionnaires for teachers and interview schedules for head teachers. The instruments were pilot-tested to ensure validity and reliability, with a test-retest reliability coefficient of 0.7 set as the minimum threshold. The study recommends strengthening peer coaching programs through structured frameworks for classroom observation and feedback, enhancing collaborative work structures by creating dedicated time for teacher collaboration, promoting action research engagement through training and institutional support, fostering professional dialogue culture through structured conversations and reflective discussions.

Keywords: peer supervision, professional learning, collaborative theory, clinical theory, Pearson’s correlation, junior secondary schools, Bungoma south sub-county

DOI https://doi.org/10.33803/j63cs5Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/j63cs5 ↗

84.Streamlining Academic Scheduling: A Web-Based Timetable System with Real-Time Conflict DetectionAbstract

John Gachie Wambui^1*⁠iD

Institutional scheduling remains a complex challenge in resource-constrained educational environments, where managing multi-variable constraints manually often leads to spatial and personnel overlaps. This project investigates a centralized digital approach to constraint satisfaction in academic timetabling. A prototype web application was engineered utilizing a Python/Django backed and a relational database architecture to model departments, instructors, and physical facilities. The core methodology involved implementing a custom, rule-based algorithmic validation framework that dynamically evaluates time slots, room capacities, and personnel availability during concurrent administrative data entry. Initial deployment and testing within a simulated administrative environment demonstrated that the rule-based validation successfully flagged and blocked targeted constraint violations. The system effectively reduced manual validation steps while maintaining data integrity across concurrent user sessions. This prototype offers a practical, applied computing solution for digital timetable management, providing a foundational framework for scalable resource optimization in educational technology.

Keywords: constraint satisfaction, applied computing, educational technology, rule-based validation, academic timetabling, relational database architecture, resource optimization

DOI https://doi.org/10.33803/ganpp7Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/ganpp7 ↗

85.A Stochastic Framework for Intermittency-Aware Ambient RF Energy Harvesting for Wireless Body Area NetworksFull paper

Dennis M. Kavita^1*, Michael D. Odhiambo², Steven O. Awino¹, Fanuel M. Keheze¹

Ambient Radio Frequency (RF) energy harvesting offers a promising solution for powering battery-less Wireless Body Area Networks (WBANs), but its practical reliability is constrained by the stochastic and intermittent nature of ambient RF sources. Existing studies largely emphasize antenna and rectifier efficiency without providing a unified approach for quantifying intermittency. This paper proposes a stochastic framework integrating RF propagation, shadowing, multipath fading, source activity, and multiband power aggregation. Two quantitative metrics are introduced: the Intermittency Mitigation Score (IMS), a weighted measure combining receiver sensitivity, number of harvested frequency bands, flexibility, and low-power conversion efficiency, and the Intermittency Robustness Factor (IRF), which measures the relative reduction in outage probability achieved through multiband harvesting. The framework is evaluated using experimentally reported performance data from representative rectenna designs in published literature. Comparative results indicate that designs combining high sensitivity, multiband operation, and efficient low-power conversion provide greater resilience to intermittent RF availability than designs optimized primarily for peak conversion efficiency. The proposed framework therefore provides a quantitative basis for evaluating and designing intermittency-aware ambient RF energy harvesters for WBAN applications.

Keywords: ambient RF energy harvesting, wireless body area networks, stochastic modeling, intermittency mitigation, multi-band rectennas, outage probability, RF-to-DC energy conversion, energy harvesting reliability

DOI https://doi.org/10.33803/bvrbraPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/bvrbra ↗

86.Beyond the Bin: Engineering a School-to-System Pathway for Circular Waste Management in KenyaFull paper

Ephraim Mwangi Kinyua^1*, Joseph Muguro¹⁠iD

The rapid growth of municipal solid waste (MSW) presents an environmental, economic, and institutional challenge for countries seeking sustainable development. The United Nations Environment Program projects global MSW generation to increase from approximately 2.3 billion tonnes in 2023 to 3.8 billion tonnes by 2050, while the economic costs associated with inadequate waste management are expected to increase substantially under business-as-usual trajectories. In Kenya, the Sustainable Waste Management Act, No. 31 of 2022 establishes source segregation requirements and places corresponding responsibilities on waste generators and service providers. However, recent empirical research in Kenyan basic education institutions indicates a substantial gap between awareness of waste management and the preservation of separated waste through subsequent collection and processing systems. This paper develops a school-to-system framework in which education is treated not as a standalone solution to waste management, but as an upstream mechanism capable of generating sustained source-separation behavior when supported by appropriate infrastructure, collection systems, markets, and resource-recovery facilities. The framework links school-based environmental education to four downstream outcomes: material recycling, organic-waste recovery, reduction in disposal demand and associated methane emissions, and energy recovery from appropriate residual waste. Evidence from Kenya and international comparative cases identifies the conditions required for these pathways to function. Nairobi waste-characterization data are also examined to demonstrate the importance of waste composition and moisture content when evaluating waste-to-energy potential. The analysis distinguishes demonstrated relationships from proposed causal pathways and engineering scenarios. It further considers the implications of source separation for Kenya’s informal recycling sector, particularly in the context of the planned transition away from the Dandora dumpsite. The paper argues that the principal opportunity is therefore not simply to “educate children about waste”, but to connect school-based behavioral change to a functioning circular-economy infrastructure in which separated materials retain economic and environmental value after leaving the school.

Keywords: circular economy, source separation, municipal solid waste, recycling, composting, resource recovery, waste-to-energy, decarbonization

DOI https://doi.org/10.33803/d9qu79Publisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/d9qu79 ↗

87.Opportunistic Precipitation Estimation via Commercial MicrowaveLinks: A Systematic Review, Taxonomy, and Empirical AssessmentAbstract

Samuel Mbugua Wambui^1*⁠iD, Ciira Maina¹⁠iD, Frankline Manene²⁠iD

Opportunistic rainfall estimation using Commercial Microwave Links (CMLs) from cellular networks offers a transformative approach for high-resolution Quantitative Precipitation Estimation (QPE), especially in complex terrain and sparse weather stations regions. Converting raw signal attenuation into path-averagedrain rates requires a four-stage processing chain: dry/wet classification, clear-sky baseline drift correction,Wet Antenna Attenuation (WAA) compensation, and path-averaged inversion. In this work, we review and synthesize two decades of foundational CML literature and evaluate several open-source pipelines againstreal-world observations using the open-access OpenRainER dataset in Northern Italy (151 commercial links at 24.55–25.61 GHz collocated with 319 ARPAE weather stations). We do comparison of the four standard classifiers—rolling standard deviation, moving-median consensus, spectral decomposition (STFT),and Hidden Markov Models (HMM)—coupled with static, exponential decay, dynamic ODE (SRB), andpower-law (Pastorek) WAA models.Our empirical results yields four key insights: First, during intense storms, all four pipelines strongly correlate with collocated gauges (r ≈ 0.70–0.98), reaching r ≥ 0.983 on long links (L = 11.83 km). Second,over a full annual cycle (8,760 hours), combining STFT spectral decomposition with the SRB dynamic ODE achieves the highest correlation (r = 0.530) and lowest error (RMSE = 0.832 mm/h), while suppressing dry-weather “ghost rain” to 49.1 mm (compared to 118.3–150.1 mm for traditional heuristics). Third,static WAA subtractions outperforms up to 44.9% of genuine rain on short links (L < 2.5 km), confirming severe inverse-length (1/L) volume bias. Finally, we demonstrate the fundamental observability limit inseparating antenna wetting from falling rain, highlighting the trade-offs between simple static offsets and continuous dynamic formulations for operational networks.

Keywords: commercial microwave links, quantitative precipitation estimation, wet antenna attenuation, dry/wet classification, opportunistic environmental sensing

DOI https://doi.org/10.33803/vmd2xuPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/vmd2xu ↗

88.Project Planning and Sustainability of Borehole Water Projects in Baringo North Sub-County, Baringo County, KenyaAbstract

Dennis Okambo Ongoro^1*, Nehemiah Kiplagat¹, James Mushori¹

The sustainability of borehole water projects is a critical challenge in dry areas, where inadequate project planning undermines water service reliability and safety. Despite investment in borehole water projects, accessibility in parts of Baringo North remains below 50%, raising concerns about their sustainability. This study examined the influence of project planning on the sustainability of borehole water projects in Baringo North by examining the community needs assessment, financial planning, technical planning, risk management, and environmental and climate resilience planning. Guided by Systems Theory, the study employed a descriptive correlational design. A census of 74 borehole projects across the five wards was undertaken, with the population of 1303 selected from project management committees and sub-committees, chiefs, ward administrators, county water officials, and liaison officials from various partners. Using Yamane’s formula, stratified proportionate, and purposive sampling, a sample of 307 was drawn. Approval was obtained from the KUREC, with research authorization granted by NACOSTI before data collection. Quantitative data were collected using structured questionnaires from 249 respondents comprising project management committees and sub-committees, while qualitative data were collected through semi-structured interviews with 11 key informants, including chiefs, ward administrators, county water officials, and partner liaison officials. Project planning recorded a mean of 3.613, indicating a high implementation level; however, environmental considerations, climate-resilient planning, and risk mitigation strategies recorded lower means. The findings indicated a positive and statistically significant relationship between project planning and sustainability (r = 0.761, p < .001), with project planning explaining 57.9% of the variation in sustainability (R² = 0.579, p < .001). The study concludes that comprehensive project planning is essential for enhancing the sustainability of borehole water projects in Baringo North Sub-County. It recommends strengthening planning through needs assessment, operation and maintenance cost planning, technical feasibility assessment, risk management, environmental protection measures, and climate-resilient project design.

Keywords: project planning, sustainability, borehole water projects, community needs assessment, financial planning, technical planning, risk planning, environmental and climate resilience planning

DOI https://doi.org/10.33803/rjmxkkPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/rjmxkk ↗

89.Spatio-Temporal Storage Dynamics of Alluvial Aquifers Using Remote Sensing Indicators: A Case Study of the Enguli Sand River, KenyaAbstract

Wanjiru Nyingi^1*⁠iD, Elizabeth Joseph Mbithe^1*

In Kenya, alluvial aquifers support domestic water supply, livestock and small-scale irrigation especially during the dry season. Despite the importance of alluvial aquifers in the ASAL regions, there remains limited understanding of the spatial and temporal variability in storage, particularly during the dry season. Rainfall is the primary source of recharge, while vegetation condition, soil moisture, water level and land surface temperature provide indirect indicators of groundwater presence and accessibility. This study investigates whether remote sensing indices Normalized Difference Vegetation Index (NDVI), Soil Moisture Index (SMI), Land Surface Temperature (LST) can be used to characterise changes in aquifer storage over time and space in the Enguli Sand River, Kenya. The analysis covered the period 2016-2025 using monthly time-series data and annual maps generated at a 10m spatial resolution for NDVI and SMI. The results showed relatively wet conditions in 2019 with extensive areas showing NDVI values of 0.25–0.50 and >0.50, and SMI values of 30–57% and >57%, particularly around the drainage network, indicating favourable vegetation and moisture conditions. Dry conditions (2021 &2023) , vegetation and soil-moisture conditions deteriorated, with NDVI values ≤0.50 and SMI values ≤57%. However, localized pockets of NDVI >0.50 and SMI >57% persisted along sections of the drainage network, suggesting localized zones of enhanced moisture availability. Validation done using four closely spaced observation boreholes (OB 1,2,3, &4) showed strong positive correlations between groundwater level and NDVI (r = 0.936) and SMI (r = 0.965), while LST exhibited a negative correlation level (r = −0.582). These findings demonstrate that the combined use of NDVI, SMI, and LST provides a useful remote sensing-based approach for characterizing spatial and temporal variations in moisture conditions and potential aquifer storage dynamics in sand river systems.

Keywords: arid and semi-arid lands, groundwater variability, remote sensing, sand rivers

DOI https://doi.org/10.33803/4jterzPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/4jterz ↗

90.Techno-Economic Application Frameworks for Battery Energy Storage Systems (BESS) in East Africa's Industrial Energy SectorsPoster

Peter Nguata Mwiti^1*, Yuri Njathi²

Background. Growing up in my hometown of Thika, an industrial hub , I have seen firsthand the energy reliability and cost issues that industrial consumers face daily. Frequent unannounced power outages and maximum demand capacity charges constrain the viability and competitiveness of the C&I sector across East Africa. This grid vulnerability and its subsequent economic toll set me out to investigate the techno-economic feasibility of integrating BESS as a solution.

Methods. I conducted this evaluation across five nations in East Africa: Kenya, Uganda, Tanzania, Rwanda, and the DRC. I categorized specific industrial consumer load profiles into spiky and flat-lined typologies. To measure economic feasibility, I developed and executed a 10-year discounted cash flow algorithm in Python. This engine evaluated the LCOE and LCOS by aggregating national utility tariffs, regulatory constraints, grid reliability indices (SAIDI/SAIFI), and battery degradation curves.

Results. Grid-to-battery TOU arbitrage proved to be sub-economic across all five nations when deployed as a standalone application, as the tariff spread failed to clear the $0.1881/kWh LCOS. Targeted peak shaving on spiky industrial loads yielded rapid capital recovery, achieving simple payback period of 2.97 years in Kenya. In a Kenyan plastics extrusion case study, utilizing BESS as an industrial UPS to prevent molten plastic solidification during unexpected grid outages provided a benefit of $680,000 annually. For mining sector in the DRC facing generation deficits, substituting diesel backup operations with hybrid Solar and BESS units generated a 10-year NPV exceeding $119 million.

Conclusions. These results demonstrate that BESS deployment is an economic necessity for mitigating demand charges and averting production downtime. However, capitalizing on these benefits requires an AI-driven EMS capable of navigating the tariff structures and load profiles within each region. This framework is directly useful to C&I operators, independent power developers, and policymakers seeking to optimize manufacturing profitability.

Keywords: battery energy storage, commercial and industrial, diesel displacement, peak shaving, time-of-use arbitrage, smart energy

DOI https://doi.org/10.33803/wtmkbhPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/wtmkbh ↗

91.Technological Adaptation as a Driver of Organizational Performance: Evidence from Waste Management Companies in Nairobi City County, KenyaAbstract

Miriam Wanjiku^1*

The rapid growth of urban populations and increasing waste generation have intensified the need for efficient and sustainable waste management services. In Nairobi City County, waste management companies face operational pressures that create a need for technological solutions to improve service delivery and organizational performance. This study examined the association between technological adaptation and organizational performance among waste management companies in Nairobi City County, Kenya. Guided by the Technology-Organization-Environment (TOE) framework, the study adopted a positivist research philosophy and descriptive correlational research design. The study targeted employees and managers of waste management companies. Of the 244 questionnaires issued, 195 were duly completed and returned, representing a response rate of 79.9%. Data was collected using structured questionnaires comprising 40 items, with eight items measuring each study construct. The instrument demonstrated excellent internal consistency, with an overall Cronbach’s alpha of 0.892 and a coefficient of 0.887 for organizational performance. Data was analyzed using descriptive statistics, Pearson correlation and simple linear regression. Descriptive findings indicated a moderately high level of technological adaptation (M = 3.66, SD = 1.22). Inferential findings established a strong positive association between technological adaptation and organizational performance (r = 0.700, p < 0.01). Regression analysis indicated that technological adaptation explained 49.0% of the variation in organizational performance (R² = 0.490), with the model being statistically significant, F (1, 193) = 184.64, p < 0.001. Technological adaptation was positively and significantly associated with organizational performance (β = 0.700, t = 13.588, p < 0.001). The study contributes context-specific empirical evidence on technology adaptation within Nairobi’s urban waste management sector through application of the TOE framework. It recommends investment in appropriate technologies, employee training, digital integration, and supportive policies to strengthen operational efficiency and sustainable waste management practices.

Keywords: technological adaptation, organizational performance, waste management, technology adoption, digitalization

DOI https://doi.org/10.33803/jz7mcnPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/jz7mcn ↗

92.Two Baselines, Two Answers: Screening Kenya's Phase 2 Green Hydrogen Options Under Competing Counterfactual AssumptionsAbstract

Elijah Masaka Mwalugho^1*, Patrick Maina Kimari¹

Kenya's Green Hydrogen Strategy and Roadmap identified fertiliser as the Phase 1 no-regret application, now settled by a 165 MW geothermal-powered ammonia facility under construction at Olkaria. Phase 2 names industrial feedstocks, transport fuels and export as successors. Any attempt to rank them meets an obstacle that is rarely made explicit. The cost of avoiding a tonne of carbon dioxide depends on what hydrogen is being compared against, and studies seldom state which comparison they used. This paper screens eleven candidate end-uses twice, once against the fossil practice a site follows today and once against the cheapest lower-carbon alternative it could actually procure, and declines to privilege either ordering. Further gates address thermodynamic substitutability, demand density, project finance risk and additionality. Inputs are classified by data provenance and propagated through 20,000 Monte Carlo draws. Baseline selection matters more than any technical or cost parameter in the model, reordering eight of the eleven applications. Declining to adjudicate does not prevent conclusions. Hydrogen direct reduced iron ranks first under both treatments, at a median of 93 USD per tonne of carbon dioxide equivalent, because Kenya lacks domestic gas and no cheaper route exists. Cement kiln substitution fails under both. Transport proves non-additive rather than expensive. On Kenya's low-carbon grid, a battery-electric vehicle abates almost as much as a hydrogen one, so the choice turns on cost. Road freight and mining haulage lose to batteries, while rail is cheaper than electrifying the corridor. An additionality threshold screens out four applications, including the highest scorer on project finance risk, because the market already funds them. For policymakers, two implications follow. Phase 2 support should narrow to applications surviving both comparisons, and any appraisal of a hydrogen project should state its comparison baseline explicitly, because that single unstated choice reorders most of the options.

Keywords: green hydrogen, power-to-X, abatement cost, counterfactual selection

DOI https://doi.org/10.33803/mgrf6qPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/mgrf6q ↗

93.Variations in Nematode Assemblages in Response to Forest DisturbanceAbstract

Julius L Leiririo^1*, Hannah Karuri^1*

Nematodes are the most abundant metazoans and are widely used as indicators of ecological disturbance. We applied high-throughput sequencing (HTS) technology to explore the nematode communities in protected and unprotected areas of Mt. Kenya forest. The protected areas had a greater abundance of Rhabdolaimus (8.2%), Alaimus (7.7%), Eumonhystera (6.1%), Paratylenchus (3.4%), Teratocephalus (2.3%), Mylonchulus (2.1%), Mesodorylaimus (2%), Metateratocephalus (2%), Miconchus (1.8%), Tylenchulus (1.4%), Diphtherophora (1.4%) and Paractinolaimus (1.2%). Aphelenchus was more abundant in the unprotected site, which also had significantly lower Shannon index, number of observed amplicon sequence variants and Faith’s phylogenetic diversity. Beta diversity analysis revealed a clear differentiation of nematode assemblages in the two sites (PERMANOVA, P = 0.03). Soil pH and Fe had a significant effect (P<0.05) on abundance of nematode genera. Aphelenchus was positively associated with high pH while Mesodorylaimus, Diphtherophora and Paratylenchus showed a positive correlation with Fe. These findings demonstrate that forest protection enhances soil nematode diversity and community stability, highlighting the importance of conservation practices in maintaining belowground biodiversity and ecosystem functioning.

Keywords: ecological indicators, ecosystem monitoring, forest health, soil biodiversity

DOI https://doi.org/10.33803/wd2fgqPublisher's page stieconference.dkut.ac.ke/10th-stie-proceedings/wd2fgq ↗