VALUE-AT-RISK DRIVEN FRAUD DETECTION FRAMEWORK WITH MACHINE LEARNING UNDER DATA IMBALANCE

The minimal presence of illicit operations within the overall data creates a significant challenge in detecting financial crime due to data imbalance. This paper delineates a revolutionary methodology for fraud detection. The integration of XGBoost, Random Forest, and neural networks enhances the accuracy of Value-at-Risk (VaR). Techniques such as cost-sensitive learning and SMOTE are utilized to address the imbalance and ensure the identification of fraudulent cases. This technique assists financial organizations in mitigating potential losses by concentrating on high-risk scam scenarios and utilizing Value at Risk theories. This technology has proven in practical studies its capacity to mitigate financial risk and accelerate scam detection. It provides an innovative, risk-aware methodology for transaction security.