Contribution-Aware Coalition Federated Learning in Edge-Assisted Healthcare Monitoring Systems

Healthcare digital transformation has driven an explosion in electronic medical data, which holds great potential for improving diagnostic precision and healthcare quality. Yet, it is confronted with the issue of data privacy sensitivity. Federated edge learning (FEL) addresses this challenge by enabling privacy-preserving collaborative global model training and integrating edge computing's near-source processing, while enhancing privacy and adapting to healthcare's resource-constrained environments. However, device mobility causes delayed and inconsistent data updates, which slows global model convergence and may prevent models from reaching the desired performance within limited training rounds. To address these complexities, We propose a coalition federated edge learning (CFEL) framework for FEL Healthcare Monitoring, with two specialized calculation strategies (optimized for large/small-scale user mobility) addressing mobility-induced data inconsistency and delayed updates in the healthcare device ecosystem. We design a tailored CFEL algorithm and defined the coalition Shapley value (CSV) to compute contributions within the framework. In small-scale user mobility scenarios, CSVs act as reliable aggregation weights, ensuring accurate reflection of each user's contribution in model updates. The simulation results demonstrate the effectiveness of our proposed algorithm by comparing it with different weight generation methods in different scenarios.