SU+ Digital Repository

SU+ is an online repository for the preservation and promotion of assorted digital content at Strathmore University

Off-Campus Access to restriced resources (including the ExamsBank) now requires registration using an @strathmore.edu email address

Authentication is NOT required for On-Campus Access to content

Photo by @inspiredimages

Communities in DSpace

Select a community to browse its collections.

Now showing 1 - 5 of 7

Conferences / Workshops / Seminars + 474
Documents and Proceedings of Conferences, Seminars, Workshops (and more) held at Strathmore University
Digital Archives 178
Assorted collections of resources covering various subject themes contributed by Faculty and Library Staff
Reports / Policies + 8
Public reports and policy documents
Research / Researchers / Publications 4269
Researcher Profiles / Conference presentations / Published research articles / Faculty and Corporate research outputs
Strathmore Heritage Collection 41
A digital chronicle of the History of the University presented through a mix of pictures, videos and digitized publications

Recent Submissions

Assessing comprehension in students by processing cognitive data
(Strathmore University, 2025) Matano, A.
Conventional evaluation techniques frequently fail to capture the intricate cognitive processes like inference-making, metacognition, and information integration—that go into comprehension. This study advances comprehension assessment by developing and implementing a system that processes cognitive data from functional Magnetic Resonance Imaging (fMRI) scans to evaluate students’ reading comprehension levels in real time. There is a lack of comprehensive knowledge regarding how these patterns are closely associated with the wider spectrum of comprehension, despite the fact that eye-tracking research has revealed useful insights regarding visual attention patterns and readers' employed comprehension methods. Leveraging the Cross- Industry Standard Process for Data Mining (CRISP-DM) framework, we utilized "The Alice Dataset" to train two deep learning models—EEGNet and ResNet—to predict comprehension scores based on neural activity patterns. The implemented system integrates a web-based interface, a FastAPI backend, and cloud storage, enabling users to upload fMRI scans and receive comprehension scores ranging from 0 to 100, categorized into five levels (e.g., 90–100: Excellent). Testing revealed ResNet’s superior performance, with a Mean Absolute Error (MAE) reducing to 1.73, compared to EEGNet’s instability, highlighting the former’s suitability for neuroimaging- based assessments. While traditional methods like multiple-choice tests fail to capture underlying cognitive processes, this system offers objective, automated insights into comprehension, addressing limitations such as cost and scalability through affordable preprocessing techniques. Despite challenges like high computational demands and EEGNet’s overfitting, the findings enhance comprehension assessment practices, contributing to cognitive science and education by providing educators with precise tools for tailoring interventions. Future work aims to refine model stability and expand to multi-modal data integration.
Retrieval Augmented Generation for automating enterprise technical queries
(Strathmore University, 2025) Nyabuti, S. K.
The growing complexity of enterprise IT systems, coupled with an increasing volume of technical support queries, continues to burden support teams, especially during high-demand periods. A significant portion of these queries are repetitive, leading to agent fatigue, delayed resolutions, and reduced customer satisfaction. This study presents a Retrieval-Augmented Generation (RAG) system for automating the resolution of enterprise technical queries. The system integrates a retrieval module that identifies relevant content from a domain-specific knowledge base and a generative module that produces contextually appropriate responses using pre-trained language models. Its performance was evaluated using Bilingual Evaluation Understudy (BLEU) and Recall-Oriented Understudy for Gisting Evaluation (ROUGE-L) metrics, alongside expert qualitative assessments. Results demonstrate that the RAG-based approach improves response quality and fluency, reduces manual workload, and accelerates resolution times. This study demonstrates the practical value of using RAG systems to automate repetitive support tasks in enterprise environments. Keywords: Natural Language Processing (NLP), Information Retrieval (IR), Retrieval-Augmented Generation (RAG), IT Support Systems, Generative AI.
Enhancing credit scoring in emerging markets: overcoming data scarcity with advanced machine learning and data augmentation techniques
(Strathmore University, 2025) Gathimba, R. W.
Credit risk assessment is essential for lending institutions, especially in data-scarce environments where limited borrower information complicates accurate risk evaluation. This study presents a robust machine learning pipeline that integrates real demographic data with synthetic financial records generated via a Conditional Tabular GAN (CTGAN) model, effectively augmenting the training dataset. Exploratory Data Analysis (EDA) revealed that debt-to-income and debt-to-savings ratios were among the most predictive features; these were log-transformed to address skewness and improve model learning. Four classification models Logistic Regression, Random Forest, Gradient Boosting and Neural Network were trained and evaluated. The Random Forest model consistently outperformed others when trained on a 75% real / 25% synthetic mixed dataset, achieving an accuracy of 75%, a macro F1-score of 0.69, and an AUC-ROC of 68.6%. To improve statistical reliability, bootstrapped confidence intervals were computed, confirming model robustness. A fairness analysis was also conducted by excluding sensitive attributes such as sex and marital status, resulting in an ethically aligned model without significant performance loss. The final Random Forest model was deployed using a Streamlit web application, enabling real-time credit scoring via a lightweight and user-friendly interface. This research demonstrates that synthetic data augmentation, combined with advanced machine learning, can enhance credit scoring in emerging markets, particularly for microfinance institutions. Future work will focus on fairness auditing, model calibration, and integration into financial infrastructure to maximize operational impact. Key Words: credit scoring, Random Forest, CTGAN, synthetic data, emerging markets, fairness, microfinance
An Adaptive telematics framework for usage-based insurance in Kenya
(Strathmore University, 2025) Kiruki, R. W.
This study develops a telematics-based risk classification model for usage-based insurance (UBI) in Kenya’s fleet management sector, addressing critical limitations of traditional models through adaptive machine learning techniques. The research employs ADASYN (Adaptive Synthetic Sampling) to resolve class imbalance. This approach enables more effective learning of risky driving patterns compared to conventional sampling methods. The study processes telemetry data—including acceleration, deceleration, RPM, and speed—from Easy Coach Limited vehicles collected between January 2022 and February 2024. After minority class augmentation via ADASYN, several machine learning models were evaluated, including logistic regression, random forests, and gradient boosting. Among these, XGBoost demonstrated optimal performance, achieving a classification accuracy of 99% and an area under the ROC curve (AUC) of 0.98. Notably, the model achieved a precision of 93% and recall of 97% for the minority (risky) class. ADASYN yielded a 27% improvement in the minority class F1-score compared to SMOTE, highlighting its effectiveness in balancing highly skewed datasets. The model was deployed on an Amazon EC2 instance within a secured Virtual Private Cloud (VPC). The instance hosts a Gradio-based interface, which generates a URL endpoint that can be embedded in external web applications. This allows seamless, real-time interaction with the model. Operational monitoring was implemented using Amazon CloudWatch for system health tracking and failure alerts. Model explainability was supported through SHAP (Shapley Additive Explanations), providing transparency into feature importance. Key applications of the model include precision-targeted premium adjustments based on risk scores, temporal risk analysis and operational optimization based on RPM and braking behaviors. This work contributes empirical evidence supporting a deployable framework for insurance technology in Kenya. Furthermore, it proposes policy guidelines for the insurance industry to support UBI adoption. Future research directions include applying federated learning to enhance model generalizability across East African fleets while addressing data privacy concerns.
Multimodal AI for clothing assistive solutions for the visually impaired
(Strathmore University, 2025) Kathure, B. M.
This study presents an Artificial Intelligence (AI) powered Image-to-Text-to-Speech (ITTS) system to enhance accessibility for visually impaired individuals in the clothing domain. Using the DeepFashion2 dataset, the Bootstrapped Language Image Pretraining (BLIP) model generated enriched captions, integrating metadata such as clothing scale, viewpoint, and category. These enriched captions were synthesized into audio using Google Text-to-Speech (gTTS), offering an accessible and descriptive experience. The system’s performance was evaluated under zero-shot and fine-tuned settings, demonstrating substantial improvements in Bilingual Evaluation Understudy (BLEU)-1 (from 0.09 to 0.19), BLEU-2 (from 0.04 to 0.07), BLEU-3 (from 0.02 to 0.04), Recall-Oriented Understudy for Gisting Evaluation (ROUGE-L) remained stable at 0.16. At the same time, Metric for Evaluation of Translation with Explicit Ordering (METEOR) improved from 0.09 to 0.13. Although Consensus-based Image Description Evaluation (CIDEr) scores remained at 0.0, the fine-tuned model excelled in generating contextually rich and descriptive captions due to metadata integration. This study highlights the potential of multimodal AI systems, whose performance was evaluated using BLEU and other standard metrics, to address accessibility challenges, providing a solution to empower visually impaired users and laying the groundwork for future innovations in inclusive design. Keywords: Multimodal AI, Assistive Reading, Digital Accessibility, Fashion Content, Image-to-Speech, Inclusive Design.