Refining Detection Mechanism of Mobile Money Fraud Using MoMTSim Platform

Mobile money financial crime evolves in various forms including account takeover fraud, refund fraud, and fake credentials, which poses challenges in measuring its overall cost using real transaction data. Current machine learning methods struggle due to outdated historical financial data that cannot be used to study emerging fraud patterns. This paper introduces MoMTSim, a virtual mobile money platform developed and calibrated using real mobile money transaction data from Sub-Saharan Africa to refine fraud detection mechanisms. Employing statistical methods, agent-based modeling, and social network analysis, MoMTSim improves fidelity by comparing real and synthetic data using the sum of squared errors (SSE) approach. Our experiments show consistent success in fraud classification using machine learning algorithms inclusive of Random Forest and XGBoost. Simpler models encompassing Logistic Regression, KNN, and Decision Trees also exhibit remarkable performance in mobile money fraud classification. This approach aids in studying new fraud scenarios and strengthening financial fraud detection mechanisms using the virtual mobile money platform.