AI-Powered Real-Time Fraud Detection System

The exponential growth of digital transactions across banking, e-commerce, UPI systems, and mobile wallets has led to a significant rise in financial fraud. Traditional rule-based systems struggle to adapt to evolving fraud patterns and highly imbalanced datasets. This research presents an AI-powered real-time fraud detection system that integrates machine learning, behavioral analytics, ONNX Runtime optimization, and a scalable backend architecture to enable fraud classification within milliseconds. The system leverages real-time feature extraction, an ONNX-optimized XGBoost model, FastAPI-based inference services, PostgreSQL transaction logging, and interactive dashboards for fraud monitoring. Experimental results demonstrate high precision, strong recall, and extremely low inference latency (