Fraud Detection Empowered by Federated Learning with Flower Framework on Amazon SageMaker AI

Shin Kong Financial Holding (Shin Kong Life) faced challenges in detecting fraud due to privacy regulations and isolated data, which limited model accuracy. They needed a way to collaboratively train models across institutions without sharing raw data to detect diverse fraud patterns while ensuring privacy and compliance with GDPR and CCPA. The solution leveraged federated learning with the Flower framework deployed on Amazon SageMaker AI. This enabled multiple institutions to jointly train shared fraud detection models while keeping data decentralized and private. Synthetic Data Vault (SDV) was used to generate synthetic data for validation to improve model generalization and fairness. SageMaker AI provided orchestration, automated monitoring, and security features including AWS IAM for scalable and compliant deployment. The results included improved fraud detection accuracy due to learning from a wider distribution of fraud patterns, reduction in false positives, enhanced collaboration among institutions without compromising data privacy, and the system’s scalable adoption by Shin Kong Life to more effectively prevent transaction fraud.