ADVANCEMENTS IN FEDERATED LEARNING FOR SECURE DATA SHARING IN FINANCIAL SERVICES

This paper explores the application of Federated Learning (FL) in the financial sector, focusing on enhancing security and privacy in key areas such as fraud detection, Anti-Money Laundering (AML) compliance, and biometric authentication systems. FL enables collaborative model training across multiple financial institutions without sharing sensitive transaction data, thereby preserving privacy while improving the accuracy of fraud detection models. In AML compliance, FL facilitates the development of robust models by leveraging diverse datasets, enhancing the ability to detect suspicious activities. Moreover, FL strengthens biometric authentication systems by decentralizing model training, reducing the risks of data breaches, and ensuring compliance with privacy regulations. The paper also evaluates the performance of a loan default prediction model trained using FL, highlighting challenges with class imbalance and model bias toward the majority class. The classification report indicates high recall (98%) but also shows a potential for misclassifying non-default cases, leading to a moderate precision (81%) and an F1-score of 89%. The model's AUC of 0.69 suggests moderate discriminatory power, with room for improvement in its ability to differentiate between default and non-default cases. The model achieves an overall accuracy of 80%. Despite these challenges, it demonstrates good generalization capabilities while maintaining the privacy of client data, presenting a promising approach to secure financial transaction analysis.