Adaptive Fraud Detection in Online Transactions Using Machine Learning Techniques

Financial institutions suffer substantial losses due to fraudulent online transactions. Traditional fraud detection methods often fail to identify evolving fraud patterns due to dataset imbalance and high false negative rates. This paper proposes an adaptive fraud detection system integrating a value- at-risk (VaR) metric with machine learning techniques. The system uses historical simulation to estimate potential fraud- related losses and employs K-Nearest Neighbors (KNN) to classify fraudulent transactions. The proposed approach enhances fraud detection accuracy while minimizing false negatives, providing an effective fraud prevention framework for financial institutions.