Comparative Analysis of UPI Fraud Detection using Machine Learning

The rapid proliferation of Unified Payments Interface (UPI) transactions in India has been accompanied by a sharp escalation in digital payment fraud, rendering conventional rule-based detection mechanisms fundamentally inadequate. This paper presents a rigorous and reproducible comparative evaluation of five supervised machine learning algorithms— Linear Regression, Logistic Regression, Support Vector Machine (SVM), Random Forest, and XGBoost—for detecting fraudulent UPI transactions within a unified experimental framework. Each model is trained and validated on an annotated dataset of 60,000 UPI transaction records, encompassing behavioral, temporal, and device-centric attributes. Class imbalance is addressed through the Synthetic Minority Oversampling Technique (SMOTE), applied strictly within training folds to prevent data leakage, while hyperparameter optimization employs stratified 5- fold cross-validation. Performance is assessed across accuracy, precision, recall, F1-score, and ROC-AUC metrics. Experimental results demonstrate that XGBoost achieves superior detection accuracy of 97.4% with an ROC-AUC of 0.99, substantially outperforming all classical approaches. A full-stack deployment architecture integrating a Next.js frontend, a FastAPI/Flask inference layer, and a Python-based ML pipeline operationalizes real-time fraud classification. To the best of the authors’ knowledge, this work constitutes one of the first studies to conduct a systematic, protocol-controlled comparison of these five algorithms exclusively on UPI transaction data, offering actionable deployment guidance for adaptive fraud prevention in large-scale digital payment ecosystems.