Theses and Dissertations
Permanent URI for this collection
Browse
Recent Submissions
Item An optimized fast learning network model for real time intrusion detection in network security(Muni University, 2025-10-22) Wamusi, RobertTraditional Intrusion Detection Systems (IDSs) often struggle with high false alarm rates, lengthy training periods, and challenges in quickly identifying emerging threats, such as DDoS and phishing attacks. At Uganda Christian University (UCU), where IDS relies on network traffic analysis, these issues can hinder early threat detection and prompt a rapid response. This research examines the current network threat landscape at UCU, focusing on its types, characteristics, and attack strategies. To address data imbalance and high dimensionality, various techniques, including configurations of the Synthetic Minority Over-sampling Technique (SMOTE), Random Forest, XGBoost, AdaBoost, Decision Trees, and Convolutional Neural Networks (CNNs), are employed. The best results were obtained with Random Forest combined with SMOTE, achieving an accuracy of 81.88%, a precision of 82.17%, a recall of 81.88%, and an F1-score of 80.19%. This model approach performed better than traditional algorithms in terms of accuracy and speed, which can be implemented in university networks.Item Development of machine learning integrated cyberbullying response framework for tertiary institutions(Muni University, 2025-10-27) Torome, Alahai AhmedWith the extensive availability of online learning and digital interactions that have arisen in tertiary institutions since the post-COVID-19 period, this study addresses the growing problem of cyberbullying among students in Arua tertiary institutions entitled as “Development of a Machine Learning Integrated Cyberbullying Response Framework for Tertiary Institutions”. The research is on Muni University, Uganda National Institute for Teacher Education (UNITE), and Arua School of Comprehensive Nursing. The study was based on Ecological Systems Theory and Social Learning Theory and described environmental influences and behaviors that lead to cyberbullying. It identifies underlying drivers including the proliferation of smartphone and social media use, absence of cyber safety policies, peer pressure, online anonymity and limited institutional responsiveness. Cyberbullying on a national level impact approximately 42% of students in Uganda, with harassment, impersonation, and identity-based attacks being the most prevalent — particularly against marginalized groups, like LGBTQ+ students. A mixed-methods design that included surveys for quantitative insights, interviews/focus group discussions for qualitative insights, and machine learning approaches such as K-Means clustering, the Elbow Method, and classifiers such as Random Forests and Decision Trees was utilized. The investigation involved 500 participants (students, staff, and administrators) selected via a multistage sampling process. Main Findings revealed that 60.9% of students had experienced cyberbullying, namely via social media. LGBTQ+ students were significantly more likely to fall victim. It showed through quantitative analysis that low awareness, limited resources, and psychological impacts were three major predictors of cyberbullying. The K-Means model classifies students by their risk levels based on both institutional backing and lived experiences, providing the foundation for an automated response mechanism. The research finds that cyberbullying is rife and more frequently affects minority groups than other classes. Institutional policy failures, lack of support and student ignorance were the major factors that contributed to the discrepancy. The machine learning inspired framework showed promising real-time risk detection potential. Propositions for the use of AI across organizational functions such as integrating the system, regular retraining, broadening the behavioral input variables, training the staff to interpret the outputs effectively and ethical use of the AI were presented. The study also suggests that research using natural language processing and long-term tracking could help strengthen interventions and protect students even betterItem A blockchain-based CCTV data verification framework for Uganda police force: a case study of Kampala metropolitan area(Muni University, 2025-10-23) Adebo, ThomasThe deployment of Closed-Circuit Television (CCTV) systems is increasingly critical to enhancing public safety and crime investigations worldwide. In Uganda, despite significant investments by the Uganda Police Force (UPF) in the Kampala Metropolitan Area, existing CCTV systems face major challenges related to centralized data storage vulnerabilities, including risks of tampering, unauthorized access, and data manipulation. These challenges undermine the evidentiary integrity of surveillance footage, negatively impacting judicial processes and public trust in law enforcement. This study aimed to design and implement a Blockchain-based CCTV data verification framework to address these issues by securing and authenticating surveillance footage. Guided by the Design Science Research Methodology (DSRM), the research involved identifying current challenges and designing a Blockchain-based Framework using Multichain Blockchain technology. This framework was developed to include smart contracts for automated verification and validated through simulation using a dataset of 35 video footage clips from Kampala Metropolitan. A test of the Blockchain-based CCTV video verification prototype demonstrated its effectiveness, successfully uploading 30 correctly tagged video footages linked to active police case files while accurately rejecting 5 tampered, unreferenced clips. This confirms the system's ability to verify and accept only valid, case-linked footage, enhancing data integrity. The findings also revealed key weaknesses in current CCTV data management: 65% reported unauthorized access, 35% raised concerns about authenticity, trust in existing verification is low (22.5%), 55% rely on manual checks, and 60% lack proper backups. These issues underscore the pressing need for Blockchain verification, automated integrity checks, secure access, and reliable storage to enhance system trust, security, and resilience. This study contributes to the application of Blockchain technology in public sector digital forensics, offering a practical and scalable solution for law enforcement agencies. The findings underscore the potential of Blockchain to enhance data integrity, foster public trust, and streamline criminal investigations. Future research should focus on integrating artificial intelligence-driven video analytics with Blockchain systems, expanding the framework’s applicability to national levels, and addressing regulatory and operational challenges for broader adoption.Item Utilizing a long short-term memory model for the prediction of refugee arrivals in Uganda(Muni University, 2025-11-26) Abusa, MichaelUganda, currently hosting over 1.95 million refugees as of August 2025 (UNHCR/OPM), faces mounting challenges in forecasting refugee influxes, particularly from South Sudan, Sudan, and the Democratic Republic of Congo. Traditional estimation methods, such as registration, mass verification, and census, are reactive, resource-intensive, and lack predictive foresight. This study examines the application of Long Short-Term Memory (LSTM) deep learning models to predict refugee arrivals using time-series data from Rhino Camp Refugee Settlement, spanning the period from 2015 to 2024. Adopting a hybrid research design, the study modeled five influx scenarios, including emergency-only, protracted-only, and mixed, to evaluate the LSTM model’s predictive performance against a linear regression baseline. The RMSE, R2, and MSE metrics were used to assess the model accuracy. Testing the four hypotheses yielded the following insights: Ho1: The availability of reliable and high-quality time-series data enabled effective predictive modeling, leading to the rejection of Ho1. Ho2: The LSTM model demonstrated superior reliability and precision, especially under protracted influx conditions, thereby rejecting Ho2. Ho3: Significant differences emerged in model performance across scenarios. The protracted-only scenario achieved the highest accuracy (MSE = 0.0191, RMSE = 0.1276), while the emergency-only scenario recorded the highest error (MSE = 0.0251), confirming the rejection of Ho3. Ho4: A significant relationship was found between refugee arrivals and influx timing, particularly in structured displacement settings, leading to the rejection of Ho4. Overall, the LSTM model consistently outperformed the linear regression baseline, capturing complex temporal dependencies in refugee arrival patterns. Training times ranged from 6 to 20 minutes, underscoring computational efficiency. The study concludes that LSTM-based forecasting, when grounded in structured and temporally stable datasets, offers a scalable and ethically sound approach to enhancing early warning systems and humanitarian planning. These findings contribute to the growing body of knowledge on AI applications in displacement contexts and offer a framework for future research and operational deployment.