Multimodal biometric authentication method by federated learning

Abstract

The Internet of Health Things requires rigid security policies to control access to sensitive data. However, nowadays, classic methods for user authentication may not meet the requirements for protection against unauthorized users during the collection, storage, and transmission of data. Therefore, there is a need for the evolution of technologies that allows the authentication of users based on unique personal identifiers (biometric characteristics). This work presents a security management approach for authentication that stands out for using two combined convolutional neural networks (CNN) for the biometric identification of users. The new approach relies on Federated Learning (FL) which is a Machine Learning paradigm that can support data management and privacy by training decentralized models collaboratively without effective data exchange. The new approach also combines Photoplethysmography and Electrocardiogram signals which improves identification accuracy and establishes a multimodal authentication. In sum, the new security management approach achieves high Accuracy, and a low false acceptance rate, guaranteeing protection against unauthorized access attempts.