COPAS http://localhost/xmlui/handle/123456789/1281 2026-09-16T02:24:27Z 2026-09-16T02:24:27Z Odongo, Winnie Gift http://localhost/xmlui/handle/123456789/7111 2026-08-31T12:56:15Z 2026-08-31T00:00:00Z

Fingerprint Classification Using Kmcg Algorithm under Varying Window and Codebook Sizes Odongo, Winnie Gift Fingerprint classification is a key task in biometric recognition because it supports faster identification, verification, and retrieval in automated fingerprint systems. However, accurate classification remains difficult when fingerprint images contain similar ridge patterns, noise, partial impressions, or variations caused by acquisition conditions. Traditional machine learning methods are efficient and interpretable, but their performance depends on the quality of handcrafted features. Deep learning models often achieve higher accuracy, but they require greater computational resources and provide limited transparency. This study evaluated fingerprint classification using Kekre’s Median Codebook Generation (KMCG) under varying window and codebook sizes and compared its performance with Principal Component Analysis (PCA) and deep learning approaches. The main objective was to determine the effectiveness of KMCG-based feature extraction and compare its performance with PCA-based machine learning and deep learning models. Fingerprint images from the NIST Special Database 302 were used. KMCG extracted texture descriptors under varying window and codebook sizes, while PCA served as a dimensionality-reduction baseline. Support Vector Machine, K-Nearest Neighbours, Random Forest, and XGBoost classifiers were trained using the extracted features. Three convolutional neural network architectures were also implemented to represent different deep learning capabilities: MobileNetV2 was selected as a lightweight and computationally efficient model, InceptionV3 was used to evaluate multiscale feature extraction, and DenseNet201 was included to assess whether dense feature reuse and greater network depth improved fingerprint classification. Performance was evaluated using accuracy, precision, recall, F1-score, average score, confusion matrices, and execution time. MobileNetV2 achieved the best overall performance, with an accuracy of 91.7% and an average score of 94.9% across the evaluation metrics. DenseNet201 achieved 90.0% accuracy, while InceptionV3 recorded 83.4%. Among the traditional machine learning approaches, PCA-XGBoost obtained the highest accuracy of 73.8%, followed by KMCG-SVM at 70.2%. These findings demonstrate that deep learning models provide greater classification accuracy, although KMCG-based approaches remain useful where computational efficiency, compact feature representation, and interpretability are important. The study therefore provides a comparative framework for selecting fingerprint-classification approaches according to performance requirements and available computing resources. MSc in Software Engineering

2026-08-31T00:00:00Z Chacha, Josephine Mweyeli http://localhost/xmlui/handle/123456789/7105 2026-08-11T11:19:47Z 2026-08-11T00:00:00Z

Adoption of Shallow Neural Networks in Pneumonia Classification Chacha, Josephine Mweyeli In low-resource healthcare environments, limited access to radiologists and a high computational burden of conventional deep learning models are significant challenges for pneumonia diagnosis. The current Convolutional Neural Networks (CNNs) for chest X ray classification demand high memory, computation and specific hardware, which is not feasible in under-resourced hospitals and clinics. In this work, the authors explored the possibility of a lightweight shallow CNN producing reliable pneumonia classification without being too computationally expensive. The architecture proposed was comprised of three convolutional layers to reduce the amount of computational and memory resources without compromising the diagnostic performance. The model was tested on a set of common experimental settings, with the other popular lightweight and deeper CNN models. To compare the performances fairly, all the models were optimized in the same way, trained by the same augmentation methods and assessed by the same evaluation metrics. It involved two data sets: the first was a secondary benchmark data set from publicly available chest X-ray repositories, and the second was a Kenyan primary data set collected from health care facilities. The accuracy, precision, recall, F1-score, and Type I and Type II error rate were used as evaluation metrics. The proposed model was found to be 91% accurate on the secondary benchmark data set and 95% accurate on the primary data set of the Kenyan. The model showed better stability of convergence, reduced overfitting and reduced majority-class bias in comparison with deeper architectures. The study proposes a framework for CNN that is efficient in terms of computation and scalable, which is suitable for resource-limited healthcare systems, where there are limited GPUs, memory, and a lack of digital infrastructure in low-resource clinical environments. Keywords: Pneumonia, Lightweight, Imaging, Resource-Constrained, Artificial Intelligence. MSc in Computer Systems

2026-08-11T00:00:00Z Buleti, Sylvia Imbuhila http://localhost/xmlui/handle/123456789/7094 2026-08-07T12:07:20Z 2026-08-07T00:00:00Z

Sustainable Intensification of Smallholder Farming Systems Using Push-Pull Technology as a Template Buleti, Sylvia Imbuhila Sustainable intensification is essential for increasing food production on shrinking smallholder farms while reducing environmental risks. In western Kenya, farmers use practices such as intercropping, crop rotation, agroforestry, crop-livestock integration, and push-pull technology to address constraints such as low soil fertility, pests, diseases, and limited land for production. Push-pull technology, which combines cereal crops with repellent and trap companion crops, is effective in managing striga weed, stem borer, and fall armyworm, but its adoption remains limited because it is mainly cereal-based and its companion crops are not edible. This study aimed to identify farmers’ preferred sustainable intensification options for integration into push-pull systems and to assess the effects of the selected option on soil fertility, pests, natural enemies, and maize productivity in Kisumu, Siaya, and Vihiga counties. Farmers’ preferred intensification practices were identified through participatory research involving focus group discussions, key informant interviews, and validation with farmers. Integration of Cajanus cajan (pigeon pea) into push-pull technology was selected as a priority option because pigeon pea provides additional benefits, including food, feed, and fuelwood. Field trials were then established on farmers’ plots during the long and short rainy seasons of 2021, 2022, and 2023 using four treatments: push-pull, push-pull with pigeon pea, maize with pigeon pea, and maize monocrop. Subplots were used to monitor crop growth, striga weed density, fall armyworm, stem borer, and natural enemies. Equally, soil physical and chemical properties, maize grain and stover yield, pigeon pea productivity, Brachiaria biomass, and desmodium biomass were measured in the same plot. Data were analyzed using analysis of variance, and treatment means were separated using Tukey’s Honestly Significant Difference test. Farmers identified several sustainable intensification options for integration into push-pull systems, with intercropping, crop rotation, crop-livestock integration, and agroforestry ranked as the most preferred. Their main motivations were food diversification, fuelwood and fodder provision, soil improvement, and pest management. Soil fertility varied across counties, ranging from low in Vihiga to moderate in Kisumu and Siaya. Major nutrients, particularly nitrogen, phosphorus, and potassium, were generally below critical levels, while micronutrients such as zinc and manganese were sufficient. Integrating pigeon pea improved nitrogen, phosphorus, and potassium levels in the tested soils, indicating its potential to support soil fertility improvement in maize-based systems. Maize monocrop recorded the highest striga density and the greatest infestation by fall armyworm and stem borer, whereas maize with pigeon pea had significantly lower striga density. Maize grain yield did not differ significantly for treatments across sites and seasons, indicating that intensified systems-maintained productivity. However, maize monocrop produced the highest stover yield, followed by push-pull, maize + pigeon pea, and push-pull + pigeon pea. Pigeon pea also contributed additional benefits, including food, fodder, and fuelwood, while supporting striga suppression and maintaining maize productivity. The findings show that integrating pigeon pea into push-pull and other maize-based farming systems can diversify farm benefits without compromising crop productivity. The study recommends the integration of pigeon pea into push-pull and maize-based systems in Kisumu, Siaya, and Vihiga counties as one of the practical strategies for improving soil fertility, suppressing striga weed, maintaining crop production, and providing additional farm products such as food, fodder, and fuelwood. PhD in Plant Science

2026-08-07T00:00:00Z Nyambura, Shalyne Gathoni http://localhost/xmlui/handle/123456789/7093 2026-08-07T08:48:43Z 2026-08-07T00:00:00Z

A Likelihood-Based Multiple Change Point Algorithm for Count Data with Allowance for Over-dispersion: A Case of COVID-19 Infections in Kenya Nyambura, Shalyne Gathoni Count data frequently exhibit over-dispersion, where the variance exceeds the mean, limiting the effectiveness of conventional changepoint detection methods that assume equi-dispersion. ThisstudyaddressesthislimitationbydevelopingahybridLikelihood Based Negative Binomial Multiple Changepoint Algorithm (NBMCPA) capable of detecting multiple changepoints in both equi-dispersed and over-dispersed count pro cesses within a unified framework. The algorithm exploits the limiting relationship between the Negative Binomial and Poisson distributions, allowing a single likelihood formulation for both data types. It integrates Stepwise Recursive Binary Segmentation, maximum likelihood estimation, and likelihood ratio testing to identify statistically significant changepoints, while Monte Carlo simulation provides critical values for reliable statistical inference. Performance is evaluated using simulated datasets with varying sample sizes and changepoint locations. Results show that the algorithm accu rately detects true changepoints with low false detection rates, with detection accuracy improving as sample size increases. Application to daily averaged COVID-19 infection data from Kenya (March 2020–August 2021) identified four statistically significant changepoints corresponding to major epidemiological developments and public health interventions. Lag analysis showed that observable changes in infection trends oc curred, on average, approximately 38 days after policy implementation, while infection peaks followed interventions by about one week. The study introduces a novel hybrid likelihood-based framework thatextendsexistingchangepointmethodologybyunifying the analysis of equi-dispersed and over-dispersed count data within a single Negative Binomial likelihood approach. The proposed algorithm provides a robust, flexible, and computationally efficient tool for detecting structural changes in count data, with applications in epidemiology, public health surveillance, environmental monitoring, finance, and industrial quality control. PhD in Applied Statistics

2026-08-07T00:00:00Z