The study compares the performance of machine learning (ML) algorithms in detecting fraud on fintech platforms, with a focus on Uganda’s mobile money ecosystem. With financial fraud evolving in complexity, traditional rule-based systems struggle to keep up. The research evaluates supervised, unsupervised, and hybrid ML approaches using a publicly available dataset, and Python-based implementations. The study results show the superior performance of ensemble methods (Random Forests) and Neural Networks (deep learning models) in detecting fraudulent activity, especially when enhanced by techniques such as Synthetic Minority Over-Sampling Technique (SMOTE) and AutoEncoders. This work provides a comparison of the performance of different ML models and reveals the significance for Uganda’s fintech firms to adopt context-sensitive, data-driven fraud detection systems. It proposes strategic recommendations for data sharing, labour capacity building, regulatory reform, and Artificial Intelligence (AI) adoption.
