Identifying UPI Fraud Through Machine Learning

Abstract

The rapid use of the Unified Payments Interface (UPI) has increased the risk of online fraud. To solve this, we propose a fraud detection method using six machine learning algorithms: Decision Tree, Random Forest, Gradient Boosting Machines (GBMs), and XGB Classifier. The Decision Tree approach offers clear decision-making processes for transaction categorization. Random Forest detects fraud more effectively by boosting robustness and accuracy. By combining weak learners to capture complex fraud patterns, GBMs detect evolving fraud tendencies over time. training the model for efficient convergence. The XGB Classifier is a powerful gradient boosting technique for classification applications. It is fast, prevents overfitting, and handles missing data. Our multi-algorithm approach enhances UPI security and ensures secure and accurate processing of UPI transactions by distinguishing fraudulent from legitimate transactions. The model is prepared for use in actual financial systems. The objective is to quickly identify suspicious or fraudulent transactions by looking at patterns in user behavior, transaction frequency, amount, location, and device-related metadata. The system employs both supervised and unsupervised learning approaches to ascertain if a transaction is authentic or fraudulent. Evaluation criteria including as precision, recall, F1-Score, and ROC-AUC are used to ensure high accuracy and low false positives. The proposed approach enhances digital payment security by proactively detecting fraud, increasing customer trust and creating a safer financial environment. This project aims to develop an intelligent fraud detection system for UPI transactions using machine learning (ML).