A Value-at-Risk-Based Machine Learning Framework for Fraud Detection in Imbalanced Financial Data

Abstract

This study delineates a machine learning strategy that is based on Value-at-Risk (VaR) and is designed to detect fraud in unbalanced financial datasets. These figures demonstrate that a negligible proportion of financial activity is comprised of illegal transactions. The identification of unusual patterns is a challenge for conventional fraud detection instruments. In large part, this is the result of unpredictable transaction behavior and class imbalance. This investigation integrates VaR analysis with Random Forest, XGBoost, and SVMs. This combination enhances the detection of suspect and high-risk transactions. The proposed method enhances the performance and classification of the model by employing feature selection, oversampling, and normalization. The method reduces false positives and negatives while detecting fraud in real time. This is accomplished through the examination of financial risk indicators, anomalies, and transaction behavior. Based on the findings, the VaR-based machine learning model outperforms conventional fraud detection methods in terms of precision, recall, accuracy, and overall risk prediction. This research is dedicated to the development of scalable and intelligent financial security solutions that will assist financial organizations in reducing fraud losses and better managing risk.