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 4277
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

Application of artificial intelligence to improve efficiency in the judiciary
(Strathmore University, 2025) Mule, D. M.
Artificial intelligence, a burgeoning force across various sectors, has found its niche in legal proceedings as well. Despite the abundance of published court decisions, the manual scrutiny required for referencing past cases often prolongs the decision-making process in court. To streamline this cumbersome task, this study aims to harness the power of Artificial Intelligence, specifically through Natural Language Processing (NLP) tools. By employing techniques such as keyword information retrieval and extractive text summarization, the objective is to facilitate the jury's access to pertinent court decisions on Kenya Law.org. This study evaluates the performance of the BERT model in a legal context, particularly focusing on its ability to classify court cases accurately. The BERT model achieved an accuracy of 66.67%, meaning it correctly predicted the outcome of approximately two out of every three cases. It demonstrated a precision of 66.67%, indicating that two-thirds of the cases predicted as "Convicted" were indeed correct. Notably, the model achieved a perfect recall of 100%, signifying that it identified all actual "Convicted" cases without missing any. Despite the lower precision, the F1 score of 80% reflects a balanced performance, emphasizing the model's strength in detecting all relevant positive cases while maintaining reasonable precision. These results suggest that BERT is an efficient tool for legal case classification, particularly in contexts where ensuring that no positive case is missed is more critical than minimizing false positives. This makes BERT a promising model for supporting legal decision-making and analysis in courts. Given the substantial backlog of cases inundating Kenyan courts, compounded by the intricate nature of these legal matters, our research has underscored the potential for AI to complement human capabilities. These technologies offer a promising avenue for automating various legal processes, thereby streamlining judicial operations.
An Explainable AI model to predict financial exclusion in Kenya
(Strathmore University, 2025) Wamalwa, L. K.
Financial exclusion remains a significant barrier to economic development and social equity, particularly in emerging economies. This study employed an explainable machine learning framework to predict financial exclusion in Kenya using nationally representative survey data from 2016 and 2021. The research followed the Knowledge Discovery in Databases (KDD) process, incorporating robust feature engineering techniques to derive behavioral and demographic indicators from raw survey data. Logistic Regression, Decision Tree, Random Forest, and Gradient Boosting were evaluated under two experimental scenarios: baseline training and synthetic minority oversampling (Synthetic Minority Oversampling Technique (SMOTE)) to address class imbalance. A temporal validation strategy was implemented by training models on the 2016 dataset and testing on 2021 data to assess generalizability over time. Feature selection using Random Forest importance and SelectFromModel was applied to reduce dimensionality, while model interpretability was achieved through SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) explainability techniques. The optimized Decision Tree model achieved the highest F1-Score (Harmonic mean of precision and recall) (F1) score of 0.926, followed closely by a soft-voting ensemble F1 of 0.906. Behavioral indicators particularly financial engagement, product category diversity, and digital finance adoption emerged as stronger predictors than demographic variables. The resulting framework not only predicted exclusion risk with high accuracy but also provided transparent, interpretable insights for policy design. The findings offered actionable recommendations to government agencies, Non-Governmental Organization (NGO)s, and financial institutions aiming to improve inclusive finance strategies in Kenya and similar socio-economic contexts.
A Model for detection of safety hazards in construction sites using convolutional neural networks
(Strathmore University, 2025) Maina, A. M.
Safety should be at the forefront of every construction endeavor as it can be a matter of life or death. As such project success depends heavily on the safety protocols enacted within a site. Safety assurance, has over the past year been under the management of safety professionals. They are responsible for risk analysis, detection of safety hazards, safety protocol compliance, housekeeping etc. This however is quite an onerous task prone to errors and omissions. For example, an individual may miss to detect exposed electrical cabling, a significant safety risk to all within the working area or vicinity. The purpose of this paper is to study the utilization of deep learning techniques in detection of safety hazards on site. The hazards in consideration were fire, electrical hazards and inhalation that can be detected visually. The research workflow constituted data collection, data preparation, model training and validation. Images were collected from various construction sites in Nairobi and augmented by additional data points from the internet. The images were annotated, prepared and augmented using Roboflow, an online annotation and image-preparation tool creating a dataset of 3000 images. The images were split in the ratio of 80:10:10 and then used to train and validate three pre-trained models namely the YOLO version 8, Faster Regional-based Convolutional Neural Network (R-CNN) and Single Shot Detector Single Shot Detector (SSD) models. A mean average precision (mAP) of 0.25, 0.16 and 0.19 was achieved for the respective models, across the three classes or hazards. These results indicated the potential for use of computer vision in safety hazard detection. The You Only look Once (YOLO) version model proved superior in terms of results and was adopted for inference in a web application through a Representational State Transfer (REST) Application Programming Interface (API). This indicated possibility of use in real world setting furthermore, the model could be deployed to a CCTV camera where it can be continually trained to improve its result output. Safety professionals and stakeholders in construction projects would then be able to identify and tackle safety risks in a shorter time span enhancing their capacity to preserve life and health while still adequately managing project resources. Keywords: Deep learning, Computer vision, Construction, Safety
Predictive maintenance of aircraft engines using machine learning: a case of the CFM56-7B26E engine
(Strathmore University, 2025) Oketch, C. M.
Understanding the Exhaust Gas Temperature margin is essential for predictive maintenance of aircraft engines. It addresses the challenges of unscheduled maintenance and equipment downtime. Since aircraft engines are expensive to purchase and maintain, airline operators strive to maximize the use of installed engines. This project used several machine learning models to predict the EGT Hot Day Margin. A real CFM56-7B26E engine dataset is used for analysis and prediction. The models were evaluated using metrics such as mean absolute error, mean square error and root mean square error. Results showed that the deep learning model performed better than all the other regression algorithms, and cross-validation against engine operational limits achieved an accuracy of 98%. SHAP analysis further identified the most relevant features influencing the model predictions. The key factors include EGT, indicated fan Speed, total air temperature, core speed, mach, and the Operational length of the engine, highlighting their significant impact on engine performance. The best-performing model was then deployed on the Streamlit community cloud platform. This project demonstrates how machine learning can enhance the reliability of aircraft engine maintenance practices. Keywords: Exhaust Gas Temperature (EGT), Predictive Maintenance, Aircraft Engine, Machine Learning.
Applying machine learning to enhance fraud detection in Kenyan digital banking
(Strathmore University, 2025) Imendi, M. E.
In Kenya, leading financial institutions have lost millions due to financial fraud. Financial fraud occurs when someone (for example, a client) loses their money or assets through deception. Despite the numerous benefits, electronic transactions have created space for malicious actors to take advantage of questionable security features to get away with financial fraud. Conventional techniques cannot address the challenges such transactions present. They are slow, costly, and inaccurate, making them unreliable in this new space. Machine Learning (ML) techniques offer hope regarding preventing these crimes from growing and wreaking havoc in the industry. They are fast, accurate, and can adapt through learning to prevent new crimes. This study investigated past fraudulent financial transactions in the Kenyan finance market, identified the attributes and features contributing to fraudulent financial transactions, and developed a reasonable approach that relies on ML techniques to detect fraudulent transactions early. The research evaluated algorithms that could assist in detecting and classifying transactions accurately, relying on datasets from the Kenya mobile banking sector. The Synthetic Minority Oversampling Technique (SMOTE) technique was used to address the data imbalance within this dataset. The dataset was split into training and test data, with feature extraction ensuring that this division was accurate and precise. Several algorithms were explored, and their performance was assessed. Before hyperparameter tuning, Random Forest achieved an Area Under the Curve (AUC) of 0.995 but failed to detect any fraudulent transactions (precision and recall for class 1 were 0.32 and 0.29, respectively), resulting in a macro F1-score of 0.63, while Logistic Regression reached a precision of 0.12 and recall of 0.65 for fraud with an overall accuracy of 73% and a macro F1-score of 0.52. After tuning, the optimized XGBoost model achieved an overall accuracy of 83%, with a fraud precision of 0.81, a recall of 0.86, and an F1-score of 0.83 for the minority class. In addition, XGBoost’s macro average F1-score improved to 0.83, and its log-loss decreased to 0.185, indicating better stability and balanced performance across classes. Adjusting the decision threshold further enhanced fraud detection, increasing recall to 0.95 with a corresponding precision of 0.74 and an F1-score of 0.83. Overall, these performance numbers confirm that XGBoost is the best-performing model for detecting fraudulent transactions in the study. Logistic regression was used to predict the outcome of events; random forest combined multiple decision trees to achieve a single result; Artificial Neural Network (ANN) assisted in recognizing patterns and solving common problems; The results showed that the algorithms efficiently and accurately detected financial fraud. Model selection was followed by training, model performance evaluation, and model tuning and optimization to enhance generalization ability. The model was validated by feeding it with actual transactions and assessing its efficacy in flagging fraud and non-fraud activities. The model was deployed behind a mobile and web application displaying the model evaluation results. Keywords: Cross-Industry Standard Process for Data Mining (CRISP-DM), XGBoost, SMOTE, ANN, Random Forest