UPI FRAUD TRANSACTION DETECTION USING MACHINELEARNING (ML)

Abstract

Unified Payments Interface (UPI) has revolutionized digital payments by enabling instant, real-time, and highly accessible peer-to-peer and merchant transactions. However, the explosive growth of UPI usage has also resulted in an unprecedented rise in fraudulent activities, including phishing, social engineering, fake payment requests, unauthorized transactions, and QR-code based scams. Traditional rule-based fraud detection methods rely on fixed thresholds and deterministic patterns, which often fail to detect evolving fraud schemes. This research presents a machine learning-based UPI Fraud Transaction Detection System designed to identify fraudulent behavior by analyzing real-time transaction characteristics, user patterns, and contextual metadata. The proposed system leverages multiple supervised ML algorithms such as Random Forest, XGBoost, Logistic Regression, and Balanced SVM, combined with data-balancing techniques like SMOTE and ADASYN to address the severe class imbalance in UPI fraud datasets. The system evaluates transaction features including transaction velocity, amount deviation, device changes, geolocation inconsistencies, beneficiary frequency, behavioral anomalies, and authentication patterns. Additionally, metadata such as transaction timestamps, UPI app usage patterns, and network fingerprints are analyzed to identify subtle indications of fraud. The experimental evaluation demonstrates that ensemble learning models outperform conventional classification algorithms, with XGBoost achieving 96% accuracy, 94% F1-score, and a 0.98 AUC score. The balancing techniques significantly improved fraud-class recall from 41% to over 92%, indicating the system’s strong capability to detect minority fraud samples. The study also implements feature importance analysis, revealing that rapid successive transactions, unusual login attempts, and sudden amount spikes are the strongest indicators of fraud. This ML-driven system is capable of realtime fraud screening, proactive alerts, and automated risk scoring of UPI transactions. It also highlights the importance of adaptive fraud detection frameworks capable of learning dynamic fraud patterns. The research contributes to the development of intelligent payment security systems and supports financial institutions and UPI service providers in mitigating digital financial crime.