Author

Description

The Unified Payments Interface (UPI) has transformed digital payments in India, processing over 13 billion transactions per month as of 2024. This exponential growth has simultaneously attracted sophisticated fraudulent activities including phishing, SIM-swap attacks, unauthorized inter-bank transfers, and social engineering-based transaction manipulation. This paper presents a comprehensive machine learning framework for real-time UPI transaction risk detection and prediction. The proposed system integrates an ensemble of gradient-boosted decision trees (XGBoost), Long Short-Term Memory (LSTM) networks for sequential behaviour modelling, and a rule-augmented anomaly detection layer.

Keywords

detection, fraud detection, LSTM, machine learning, NPCI, UPI, XGBoost.