Maximizing Utility and Quality in Smart Healthcare with Incentive Driven Hierarchical Federated Learning

Abstract

Recent development of federated learning based smart healthcare techniques facilitates distributed processing of patient data while protecting privacy. However, two significant challenges can notably degrade model learning performance. Firstly, patients often show reluctance to participate in model training due to a lack of incentives. Secondly, the quality of updated models can change significantly due to different aspects such as the size and training data quality. The incorporation of low-quality updates can subsequently degrade global model quality in FL. Considering the above scenarios, this paper formulates a utility maximization problem keeping model quality function and budget into deliberation. To solve the formulated problem, this paper proposes association algorithm utilizing Utility Maximization and Payment (UMaxP) algorithm based on matching approach to select Wireless Body Area Networks (WBANs) and hospitals in a way that increases the utility and improves model quality using Hierarchical FL (HFL). In UMaxP, the allocation of patients to hospitals is optimized to maximize both the sum of local model qualities and the overall utility of the patients. The effectiveness of the proposed algorithmic model is demonstrated through theoretical analysis, experimental validation, and comparative evaluation with various methods using real-world data. © 2025 IEEE.