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 4328
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

Automated feature engineering tool for fraud detection in financial transactions using deep learning
(Strathmore University, 2025) Buoro, S. O.
Financial fraud has become increasingly prevalent in modern corporate environments. It involves the deliberate use of deceptive tactics to achieve monetary benefits within various corporations and organizations. Conventional approaches such as manual verifications and inspections, although intended to detect fraudulent activities, frequently demonstrate shortcomings in terms of accuracy, cost-effectiveness, and efficiency. Existing automated fraud detection systems also have limitations. These includes inefficiencies, high costs of implementation, data imbalance, concept drift, false positives and negatives, limited generalizability, and difficulties with real-time processing. The quick and timely detection of fraudulent activities allows financial institutions to mitigate fraudulent conduct before it could lead to financial loss. This research developed an automated feature engineering tool for fraudulent detection. The developed solution involved utilizing deep learning (DL) techniques to analyse transactional data, thus revealing hidden trends that could indicate fraudulent activities. The developed MLP Classifier achieved an accuracy of 99.75%, surpassing the Logistic Regression and Decision Tree models. The model achieved perfect classification, with no errors in predicting fraud or non-fraudulent transactions. The significance of implementing effective fraud detection systems cannot be emphasized, as they serve as protectors of the security and integrity of the financial ecosystem. By providing protection to both financial institutions and cardholders against potential financial instability, these systems strengthen the fundamental basis of confidence on which transactions rely. Key Words: Automated Feature Engineering, Deep Learning, Financial Fraud Detection, Machine Learning.
Deep learning tool for higher learning scholarship award decisions: a case of higher institutions of learning
(Strathmore University, 2025) Nyambura, T. M.
In today’s highly competitive higher education environment, scholarship programs play a crucial role in attracting and retaining talented students, especially those from disadvantaged backgrounds. However, traditional scholarship award processes often rely on manual and subjective methods, which can lead to inefficiencies, biases, and inequitable distribution of financial aid. The lack of standardized criteria and the complexity of handling large volumes of applications further complicates the decision-making process. This study developed a deep learning-based tool to assist in higher education scholarship award decisions, utilizing convolutional neural network (CNN) to automate the selection process. Through a data-driven approach, the tool was designed to analyse student profiles across multiple dimensions, including academic performance, financial need, extracurricular involvement, and personal background. The deep learning model aims to provide more accurate, objective, and fair scholarship allocations by removing human biases and ensuring that all eligible students are assessed equitably. The developed tool was tested on a dataset of historical scholarship applications and awards, and the performance of the CNN model was evaluated based on accuracy, fairness, and efficiency compared to traditional manual methods. The model achieved an accuracy of 93% surpassing the performance of all the reviewed models at the end of fine-tuning. The findings of this study offer valuable insights into the feasibility and effectiveness of integrating AI-driven tools into scholarship management systems in higher education institutions. Ultimately, this study contributes to ongoing efforts to make scholarship distribution more equitable and aligned with institutional goals and societal needs. Keywords: Scholarship Allocation, Deep Learning, Convolutional neural networks (CNNs), machine learning, higher education, automated decision-making, data-driven scholarship distribution, AI in education
A Distributed computing prototype for climate change impact simulation: case of Nairobi, Kenya
(Strathmore University, 2025) Makungu, C.
Climate change is one of the most pressing issues facing the world in the 21st century, and Nairobi is no exception. Climate change in Nairobi impacts climate-sensitive variables, including unpredictable rainfall, rising temperatures, and increased frequencies of extreme temperature and precipitation events, significantly affecting current and future urban infrastructures, public health, transportation systems, and livelihood activities. This project sought to develop a distributed computing environment for detailed climate change modeling and effects assessment in Nairobi, Kenya. The prototype successfully processed climate data at 1km² resolution compared to previous 10-25km² models for Nairobi. Performance testing showed a 14x improvement in simulation speed over traditional systems. The system captured urban heat island effects with 92% accuracy when validated against historical weather station data. Temperature predictions achieved ±0.4°C accuracy with 95% confidence intervals. Stakeholder validation confirmed the system's practical utility with a usability score of 78.3/100. The prototype directly informed three real urban planning scenarios for Nairobi County's climate adaptation strategies. Capacity building and knowledge transfer were in focus through the training modules and workshops for the local stakeholders involved. The approach to development was carried out in cycles with the desire for changes being made based on feedback sought and which priorities there was. The study was done using an integrated approach to the collation of data: Methods of data collection included questionnaires, personal interviews, focus group discussions, and observation. The quantitative and qualitative data analysis techniques generated useful information for application in urban planning and policy. It hence improved the knowledge of climate conditions in Nairobi with concrete recommendations for climate change adaptation. Keywords: Climate change modeling, Distributed computing, Urban planning, Nairobi, High-resolution climate data, Regional climate model, Urban heat island, Data assimilation, Capacity building, Climate adaptation.
Optimizing insurance time for claim resolution through machine learning
(Strathmore University, 2025) Kiprono, G. J.
As insurance plays a key role in financial protection, the ability to predict claim resolution time has become essential for improving operational efficiency and customer satisfaction. This study applied machine learning and explainable AI techniques to estimate the time taken to resolve insurance claims using a dataset from the Association of Kenya Insurers (AKI), comprising 13,000 records with 33 features. After preprocessing and feature engineering, several regression models were trained and evaluated, with XGBoost emerging as the best-performing model—achieving an R2 of 51.8%, outperforming Random Forest (45.7%) and Gradient Boosting (46.6%), and recording the lowest RMSE and MAE values. SHAP analysis highlighted key predictors, including PolicyUpgradesLastYear, PolicyStartYear, and PolicyDurationMonths. The final model was deployed via a Streamlit interface, offering a practical and interpretable solution for real-time prediction of claim resolution time, with potential to enhance decision-making and service delivery in the insurance sector.
Enhancing loan portfolio management through multi-class classification of credit risk: a case of Kenyan financial institutions
(Strathmore University, 2025) Macharia, C. N.
The effective management of loan portfolios and credit risk is crucial for the financial stability of lending institutions. However, recent economic challenges in Kenya have heightened loan default rates, underscoring the need for improved credit risk assessment processes. Traditional methods are increasingly inadequate in addressing evolving market dynamics, prompting some lenders to explore advanced techniques such as machine learning and predictive analytics. Despite their potential benefits, the adoption of these advanced techniques remains limited, particularly among smaller financial institutions. In response to these challenges, this study developed a predictive tool for multi-class loan classification to enhance credit risk assessment and loan portfolio management. Several machine learning algorithms were compared, with XGBoost emerging as the most effective model. The study also evaluated the use of the Synthetic Minority Oversampling Technique (SMOTE) to address data imbalance, which improved classification of minority risk categories. The proposed prediction tool aligns with regulatory guidelines and offers a practical solution for lenders to strengthen credit risk monitoring and decision-making, contributing to the resilience and sustainability of financial institutions in Kenya. Keywords: Credit Risk, Loan Portfolios, Lending Institutions, Loan Default Rates, Machine Learning, Predictive Analytics, Multi-Class Loan Classification, XGBoost, Synthetic Minority Oversampling Technique (SMOTE), Data Imbalance, Decision Making, Kenya