Advanced UPI Fraud Detection Using Machine Learning

Abstract: The rapid adoption of Unified Payment Interface (UPI) systems in India has significantly transformed digital financial transactions, but this growth has also been accompanied by a sharp rise in sophisticated fraud activities. Traditional rule-based and static security mechanisms are increasingly inadequate in detecting evolving fraud patterns, as they lack the ability to analyze transaction behavior dynamically. This study proposes a machine learning–based framework for real-time UPI fraud detection that leverages transaction-level behavioral features to identify suspicious activities before financial loss occurs. The framework utilizes supervised classification models, specifically Decision Tree and Extreme Gradient Boosting (XGBoost), to analyze transaction attributes such as transaction amount, sender and receiver virtual payment addresses (VPAs), device identifiers, transaction timing, and usage frequency. A Streamlit-based interactive interface is developed to support both single-transaction verification and bulk transaction analysis through CSV uploads, enabling scalable and user-friendly fraud assessment. Model performance is evaluated using standard metrics including accuracy, precision, recall, and F1-score. Experimental results demonstrate that the proposed machine learning approach achieves significantly higher detection accuracy and adaptability compared to conventional rule-based systems, particularly in identifying anomalous user behavior patterns. The findings highlight the effectiveness of data-driven fraud detection frameworks in enhancing the security, reliability, and trustworthiness of digital payment ecosystems, offering a practical solution for mitigating UPI fraud in real-world scenarios.