Enhancing Privacy Preserving Federated Learning Using Differential Privacy

Abstract

Artificial Intelligence (AI) and Machine Learning (ML) play a crucial role in credit risk assessment but pose significant data privacy risks due to centralized data storage. Traditional ML models require financial institutions to share sensitive customer data, raising concerns about security breaches and regulatory compliance. Federated Learning (FL) offers a privacy-preserving alternative by enabling collaborative model training without exposing raw data. Additionally, Differential Privacy (DP) enhances FL’s security by adding mathematical noise to model updates, preventing data reconstruction and ensuring robust privacy protection. This study explores the application of FL, integrated with DP, for credit risk prediction using dataset. Our implementation demonstrates that FL with DP maintains comparable accuracy to centralized ML while improving data security and regulatory compliance. We also discuss key challenges, including communication costs, heterogeneous data distributions, and security threats, along with future advancements in privacy-preserving AI. This research highlights FL’s potential in financial applications, ensuring secure and fair credit risk assessment.