Studying The Effectiveness of Federated Learning in The Healthcare Domain

Thumbnail Image



Journal Title

Journal ISSN

Volume Title


University of M'sila


Federated Learning (FL) has emerged as a promising approach for training Machine Learning (ML) models in decentralized settings, such as Smart Healthcare Systems (SHSs). Adoption of FL in SHSs provides many advantages especially the preservation of patient’s privacy. This project aims to investigate the effectiveness of FL in the context of SHSs by comparing it with the traditional data-centralized approach. Specifically, four FL strategies are implemented using the Flower framework, with each strategy trained on four different types of datasets: brain tumor, pneumonia, Alzheimer's disease, and COVID-19. The performance of these FL strategies is then compared to that of a data-centralized model trained on the same datasets. Our obtained experimental results reveal that the accuracy of the FL models is in general less than the data-centralized model for all four strategies. Such disparity in accuracy is explained by the challenges associated with decentralized FL learning in SHS settings. Nevertheless, such a small extent of difference can be tolerated considering the benefits of FL brought to SHSs. Thus, we may conclude that there should be a trade-off between high accuracy and FL advantages.



Machine Learning (ML), Federated Learning (FL), Smart Healthcare Systems (SHSs), Data Privacy, Decentralized Learning, Models Aggregation Methods.