Detection of Anomalies in Unified Payment Interface Transactions Using Machine Learning

Abstract

The exponential growth of UPI transactions has introduced sophisticated fraudulent activities threatening digital payment security. Traditional rule-based detection systems fail to handle evolving fraud patterns and produce excessive false alerts. This paper presents Secure UPI, a hybrid supervised learning system using a stacking approach that combines SVM, Random Forest, XGBoost, and KNN algorithms for improved fraud detection accuracy. The system preprocesses UPI transaction data by extracting features including transaction amount, frequency, and location. Chi-square feature selection identifies the most relevant fraud indicators. The stacking meta-learner achieves 96.1% accuracy with 0.95 F1-score on an imbalanced dataset, outperforming individual classifiers. Real-time monitoring detects suspicious transactions, generates instant alerts, and enables preventive actions. The system demonstrates effective machine learning application for UPI fraud detection and prevention.