Federated Learning Frameworks for Financial Security with Encrypted Model Aggregation

Abstract

The rapid digitization of financial services has increased exposure to cyber threats, fraud, and data breaches. Conventional centralized machine learning approaches for fraud detection require aggregating sensitive financial data into a single repository, creating privacy, compliance, and security risks. Federated Learning (FL) provides a decentralized alternative, enabling collaborative model training without sharing raw data. This paper proposes a secure federated learning framework with encrypted model aggregation tailored for financial security applications. The architecture integrates secure aggregation protocols, differential privacy, and adaptive anomaly detection models. Experimental evaluation demonstrates that federated learning achieves comparable fraud detection accuracy to centralized models while significantly enhancing data confidentiality and regulatory compliance. The framework provides a scalable and privacy-preserving solution for next-generation FinTech security systems.