Mobile payment fraud detection in UPIs through machine learning techniques: A systematic review

The Unified Payment Interface (UPI) has transformed the digital payments landscape in India by providing a method for safe, instantaneous, and user-friendly financial transfer. With some arguably limited oversight, UPI was adopted quickly, but unfortunately, along with its growth, the landscape is littered with new forms of advanced fraud. Therefore, this study presents a systematic analysis of UPI fraud from 2016-2025, detailing common forms of fraud such as fake mobile apps, phishing, QR code attacks, pony payment requests, and fraud involving KYC regulation. Additionally, this paper explores advanced machine learning and deep learning models for fraud detection. Some of the models we explored include convolutional neural networks (CNNs), long short-term memory (LSTM) neural networks, and detection accuracies of up to 99.74%. Other effective models include support vector machines (SVMs), random forests, and XGBoost since they are able to classify flagged transactions as nonauthorized with high levels of precision. We see that augmented and integrated systems with biometric authentication and real-time monitoring and reporting become a layered defense against new threats to fraud. The study also highlighted the rise in autopay fraud to drive the need for continuous innovation in fraud detection, undermining those who have been manipulating UPIs and digital payments since 2019. There is a need to further incorporate supervised ML, deep learning, and gradient boosting with existing banks' processes for recovery and detection tools for malpositioning to further the financial safety of UPI users and develop a competent payment system for UPIs and their clients.