FedUPI: Federated Learning Empowered Detection of Fraudulent UPI Transactions

The Unified Payments Interface has transformed digital transactions in India, providing a fast and seamless payment experience. However, the increasing volume of transactions has led to a surge in fraudulent activities, necessitating more advanced fraud detection mechanisms. Traditional fraud detection systems rely on centralized models that require banks to share sensitive transaction data, posing significant privacy risks and leading to high false positives. To address these challenges, this project proposes FedUPI, a Federated Learning based fraud detection framework that enhances security while preserving user privacy. FedUPI enables banks and payment gateways to train fraud detection models locally without sharing raw transaction data. Instead, only encrypted model updates are exchanged and aggregated using Federated Averaging (FedAvg) to create a robust global model. This approach ensures privacy compliance while continuously adapting to emerging fraud patterns. The system provides effective fraud detection, significantly reducing false positives while maintaining transaction efficiency. Its decentralized architecture eliminates single points of failure and enhances scalability, making it adaptable to the evolving landscape of digital payments. By fostering trust in UPI transactions and promoting regulatory compliance, FedUPI contributes to a more secure and reliable financial ecosystem. With its innovative application of Federated Learning, privacy-preserving techniques, and AI-driven fraud prevention, FedUPI represents a groundbreaking advancement in digital payment security, ensuring safer and more efficient transactions.