BHMFL: Autonomous Federated Learning for Privacy-Aware Health Monitoring via Self-Balancing Data Synthesis in Smart 6G-HIoT Systems

Tianxu Yang, Jing Wang

Abstract: Smart 6G-enabled Health Internet of Things (6G-HIoT) systems represent the next generation of autonomous healthcare monitoring, leveraging ultra-reliable low-latency communications, massive machine-type connectivity, and intelligent edge computing to enable real-time health analytics across distributed medical environments. However, meeting data privacy requirements and having an effective system still needs to be solved. Existing methods need to improve on data imbalance and data distribution discrepancies across monitoring devices, making the vision of privacy-enhanced health monitoring systems unattainable. This article proposes BHMFL, an autonomous federated learning framework that addresses the challenges through scaling or balancing a certain dataset in smart 6G-HIoT systems. The framework proposes new health data formation cognate to the frequency of occurrence of different disease states through combined condition-balanced sampling and privacy data-generating technology able to keep data provided from breach. This ensures that while monitoring devices enhance the automatic representation of rare health conditions, the privacy of the devices is maintained. The framework utilizes an intelligent approach to model training that accounts for physiological and time relevance factors. Experiment results demonstrate the high performance of the proposed BHMFL framework in terms of accuracy. Additionally, the BHMFL framework has increased rare condition representation from less than 5% to around 20% without fracturing the performance across different dimensionality. These results support the intuition that BHMFL has the potential to solve the scalability and security problems of distributed health monitoring autonomously and thus can be considered beneficial for future smart 6G-HIoT systems.

External IDs:doi:10.1109/jiot.2025.3611403