UPI Transaction Fraud Detection Using Machine Learning

Unified Payments Interface (UPI) has rapidly become the backbone of digital transactions in India, enabling fast, seamless, and real-time fund transfers. However, the rise in transaction volume has also led to a significant increase in fraudulent activities, including phishing, social engineering, and unauthorized transactions. This paper presents a data-driven machine learning approach to detect UPI transaction fraud with high accuracy and interpretability. The proposed system analyzes key transaction attributes and user behavior patterns, using a LightGBM classifier to identify anomalies indicative of fraudulent activity. To enhance trust and transparency, SHAP (Shapley Additive Explanations) is integrated to provide feature-level explanations for each prediction. The system architecture includes a user interface for inputting transaction details, a Flask-based API layer, and a machine learning model trained on imbalanced financial datasets. Adaptive sampling and feature engineering techniques are employed to improve detection performance. Experimental results demonstrate that the model effectively distinguishes legitimate transactions from fraudulent ones, offering real-time prediction capability suitable for deployment in digital banking environments. This work contributes to the growing need for secure digital payment infrastructures by presenting a scalable, explainable, and efficient fraud detection solution tailored for UPI systems. *