AI Driven Fraud Detection in Financial Ecosystems: A Hybrid Machine Learning Framework

Mobile money services have emerged as a revolutionary financial technology, particularly in developing economies, providing financial access to previously unbanked populations. However, this digital transformation has simultaneously exposed financial systems to increasingly sophisticated fraud mechanisms, making fraud detection a critical challenge for financial institutions worldwide. In this study, we focused on developing a comprehensive fraud detection framework for mobile money transactions using state-of-the-art machine learning techniques. The research utilized a synthetic mobile money transaction dataset derived from PaySim, a simulator aggregating real-world financial logs from a mobile money service in an African country. We implemented and evaluated three prominent machine learning algorithms: XGBoost, Logistic Regression, and Random Forest. The performance of three machine learning models was rigorously evaluated for fraud detection: (i) the Random Forest Classifier emerged as the most robust model, achieving an exceptional Area Under the Precision-Recall Curve (AUPRC) of 0.9998. (ii) Feature engineering techniques, particularly the introduction of error balance calculations, significantly enhanced the models’ fraud detection capabilities. The development of advanced fraud detection methodologies carries significant societal implications. Through continuing to refine and advance fraud detection techniques, researchers and practitioners can play a crucial role in creating more secure and reliable financial transaction systems.