AbstractThe digital payment landscape in India has undergone a paradigm shift with the advent and widespread adoption of the Unified Payments Interface (UPI). While UPI has democratized access to cashless financial services, its exponential growth has been accompanied by a surge in sophisticated fraudulent activities, ranging from high-value transaction manipulation to unauthorized device usage. Traditional rule-based detection mechanisms often fail to adapt to these evolving threat vectors, resulting in high rates of false negatives. This paper proposes a robust machine learning framework leveraging the XGBoost classification algorithm to detect fraudulent UPI transactions in real time. The system integrates diverse attributes, including transaction velocity, device novelty, and merchant risk profiles, and utilizes Scikit-learn’s Standard Scaler for feature normalization. Deployed via a Flask-based web application, the model provides instantaneous fraud probability scores. Theoretical analysis and methodological design suggest that this approach offers a scalable, high-confidence solution for securing the modern digital payment ecosystem against dynamic financial threats.
J et al. (Mon,) studied this question.