Enhancing Digital Payment Security: UPI Fraud Detection with Advanced Machine Learning Algorithms

Unified Payments Interface (UPI) fraud poses a significant risk to financial security, since traditional fraud detection techniques find it challenging to adjust to evolving fraud trends. Current machine learning (ML) or rule-based methodologies demonstrate elevated false positive rates, lack of flexibility, and delayed detection, resulting in significant financial losses. This study presents a hybrid model (DQN-XGBoost) that integrates reinforcement learning (RL) with Deep Q-Networks (DQN) for adaptive learning and XGBoost for accurate categorisation. The DQN component dynamically optimises fraud detection strategies, while XGBoost improves classification accuracy through optimum feature selection. Experimental findings from a Kaggle-sourced dataset reveal an accuracy of 98.7%, precision of 97.9%, recall of 98.5%, and a minimal false positive rate of 1.2%, dramatically surpassing the performance of Random Forest, SVM, and Logistic Regression. This research surpasses current fraud detection methods by delivering real-time performance (45ms delay per transaction), enhanced accuracy, less false positives, and greater flexibility. The research enhances safe digital transactions by providing a scalable and intelligent framework for fraud detection.