Privacy-Aware AI in Financial Fraud Detection Using Federated Data Lakes

Abstract

Privacy preservation has emerged as a paramount concern in the deployment of artificial intelligence (AI) systems for financial fraud detection. As financial institutions increasingly rely on machine learning models trained on vast amounts of sensitive customer transaction data, the risk of data exposure—whether through centralized data breaches, insider misuse, or model inversion attacks—has grown commensurately. This manuscript presents a comprehensive, privacy-aware AI framework that integrates federated learning within a federated data lake architecture to detect fraudulent financial activities while ensuring that raw transaction data never leaves its originating institution. We begin by outlining the operational challenges faced by financial consortia in collaborative fraud detection, including regulatory compliance under GDPR, PCI DSS, and similar frameworks. We then detail our federated data lake deployment, leveraging Apache Iceberg for unified data cataloging and Presto SQL for seamless cross-node querying, combined with secure aggregation protocols that cryptographically shield client model updates.