Federated Learning in Financial Data Privacy: A Secure AI Framework for Banking Applications

Le résumé fourni par la source

Data and privacy regulation have become crucial points of concern for the financial sector due to the continuously growing innovations. As financial institutions advance their implementation of AI technology in fraud detection, credit risk analysis, and regulatory compliance, centralized machine learning affects the actualization of this goal, carrying with it some disadvantages that are This paper develops a secure FL solution for banking applications with the focus in order to provide a conceptual architecture to perform collaborative model training across the participating institutions without transferring the raw data set. It also utilizes differential privacy, secure multi-party computation, and homomorphic encryption to offer compliance with privacy laws such as GDPR and CCPA. So, to incorporate this, a threat model is described with regard to threats posed by insiders and external parties and the possibility of data leakage. There are enhanced clients, a secure Aggregator, and a Central Coordinator through which communication happens with efficient protocols included. Synthetic and real-world financial datasets are used in two practical application areas to evaluate the FL model's performance, including AML and credit risk. The results obtained from this study reveal that the developed FL models outcompete the centralized and local models by 17.9% for clients with different powers in terms of accuracy. There are zero raw data leakage concerns, and capacity tests have established the model’s capability to run over 100 clients. The findings, therefore, endorse a view that FL offers a feasible, secure, and compliant means for applying AI in the financial industry

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.