TrustPay –“ Ensuring Safe Digital Transactions via Machine Learning ”

Abstract

Traditional rule-based defenses are being challenged by an increase in fraud caused by the widespread use of India's Unified Payments Interface (UPI). We examine machine learning (ML) solutions that use transaction histories to identify fraud in real time. These systems train classifiers such as Random Forest, AdaBoost, and Gradient Boosting using rich transaction data (timestamps, amounts, account IDs, device/location information) and class-balancing techniques. High detection rates have been reported in studies; for instance, ensemble methods have reached approximately 99% accuracy, and a Random Forest model achieved approximately 96% accuracy (97% precision, 91% recall). The top-performing models are used to quickly identify fraud in streaming applications (typically through web interfaces). Overall, security is significantly enhanced by ML-based UPI fraud detection, which adaptively detects new attacks while keeping false alarms to a minimum