Money Laundering Detection in Imbalanced E-wallet Transactions with Threshold Optimization

Abstract

The swift embrace of technology within the financial industry has driven the extensive utilization of electronic payment systems, providing smooth money transfers and substituting outdated paper-based procedures. Consequently, mobile payment systems centered on electronic wallets (e-wallets) have signifi-cantly transformed contemporary finance, introducing improved security measures and transaction functionalities. However, the prevalence of unlawful activities, including money laundering and associated fraud via e-wallets, presents a substantial threat to the integrity of the financial sector. This research paper delves into the pivotal role of machine learning models in identifying money laundering activities within e-wallet transactions. The study focuses on addressing the imbalance inherent in the PaySim dataset through the oversampling technique. Employing three distinct models - Logistic Regression (LR), Gradient Boosting, and XGBoost - the research systematically evaluates their effectiveness. Notably, XGBoost emerged as the standout performer, showcasing exceptional accuracy at 99.88%, precision at 0.9984, and sensitivity at 0.999. Furthermore, a threshold moving technique is employed to enhance the model's efficiency, optimizing its performance in detecting potential instances of money laundering within e-wallet transactions. © 2024 IEEE.