Enhancing health status prediction and data security using transformer-based deep learning architectures

Abstract

This paper proposes a privacy-preserving transformer-based federated learning (PPTFL) framework designed to enhance privacy, accuracy, and computational efficiency in healthcare data analysis. Federated learning (FL) has emerged as a promising solution for distributed machine learning while preserving data privacy, especially in sensitive sectors like healthcare. However, challenges such as maintaining high accuracy and managing communication overhead remain. The proposed PPTFL framework leverages the power of transformer models to improve the performance of federated learning while integrating privacy-preserving techniques. The model demonstrates superior performance with an accuracy of 92.87%, an F1 score of 92.37%, and a privacy budget (ϵ) of 1.6, outperforming existing approaches in terms of both privacy and accuracy. The model also exhibits computational efficiency, with lower communication cost and reasonable training time. Comparative evaluations with four relevant literature models further validate the effectiveness of the proposed PPTFL framework. This work highlights the potential of PPTFL to revolutionize healthcare informatics by providing secure, accurate, and efficient solutions for federated learning applications.