Article Published In Vol.8 (March-April-2020)

Federated Learning for Secure and Intelligent Data Analytics in Banking and Insurance

Author : Jyothi Bobba and Aravindhan Kurunthachalam

The increasing volume of financial transactions in banking and insurance sectors necessitates advanced fraud detection techniques while ensuring data privacy and security. Traditional centralized machine learning approaches pose significant risks related to data breaches, regulatory non-compliance, and lack of interpretability. To address these challenges, this paper proposes a Federated Learning (FL)-based framework for secure and intelligent data analytics in financial services. FL enables multiple institutions to collaboratively train fraud detection models without sharing raw data, preserving privacy and compliance with regulations such as GDPR and PCI-DSS.The proposed methodology applies Min-Max Normalization for data preprocessing to ensure balanced feature scaling, enhancing model convergence and performance. Secure aggregation techniques, including homomorphic encryption and secure multiparty computation, are incorporated to prevent data leakage and strengthen model robustness. Additionally, Shapley values are utilized to improve model interpretability, ensuring transparency in fraud detection and financial risk assessment. The framework is evaluated against traditional machine learning and centralized learning models, considering key performance metrics such as accuracy, efficiency, security, and computational cost. Experimental results demonstrate that the FL-based approach significantly enhances fraud detection accuracy while reducing privacy risks and maintaining regulatory compliance. Moreover, the proposed model achieves improved scalability and adaptability to evolving financial threats. This research demonstrates that the FL-based approach significantly enhances fraud detection accuracy while ensuring privacy and compliance. The accuracy results indicate improvements, with Traditional ML achieving 85.4%, Centralized ML at 88.7%, and Federated Learning reaching 92.3%, highlighting its effectiveness in financial security applications.

Keywords: Federated Learning, Fraud Detection, Secure Aggregation, Homomorphic Encryption, Shapley Values.