Federated Learning-based Privacy Protection Methods in Smart Healthcare Systems

Abstract

Smart healthcare systems offer substantial advantages regarding quick and precise medical decision-making. Working with personal health data brings new privacy and security challenges that require attention from the cybersecurity community. Clients would lose control over their data when employing centralised machine learning algorithms for data analysis. By handling data at the client end and sharing the parameters to a central server, federated learning (FL) offers data privacy while addressing some drawbacks of centralised machine learning algorithms. Attacks like free-riding, inference, and man-in-the-middle, direct manipulation of central server could compromise the potential risk of losing privacy for data, which may have significant impacts. This thesis significantly contributes new approaches to federated learning-based privacy protection in smart healthcare systems to enable data analytics and improve data privacy at the network’s edge. It also presents effective solutions to huge traffic congestion issues resulting from the aggregation and broadcasting of client data models, which could cost the whole system energy, time, and accuracy. The first contribution provides a three-fold federated edge aggregator that provides privacy protection for Smart Healthcare Systems at the edge. Artificial noise functions and an iteration-based Conventional Neural Network (CNN) model are used to balance privacy protection and model performance. The second contribution presents a Serverless Privacy Edge Intelligence-based Federated Learning (SPEI-FL) architecture to address problems of client adaption and removal without impacting the overall learning processes and compromising client privacy including authentication technique using serverless computing. The third contribution proposes an effective Privacy Embedded Learning (PEL) method to addresses how machine learning models handle privacy issues by safeguarding privacy at the patient’s end, at a medical server, and in communication channels. The proposed federated learning-based privacy protection techniques achieved high performances compared with compelling tech niques and different datasets aim for smart healthcare systems, with safe monitor services and privacy-preserving paradigms.