FDBSCAN-MBKSched: A Hybrid Edge-Cloud Clustering and Energy-Aware Federated Learning Framework with Adaptive Update Scheduling for Healthcare IoT

Abstract

The explosive growth of the Internet of Medical Things (IoMT) has created huge, diverse, and noisy health data streams that require processing in real time under stringent energy and latency budgets. Conventional fuzzy clustering and synchronous federated learning methodologies tend to be plagued by noise sensitivity, excessive communication overhead, and poor model convergence efficiency. To address above mentioned issues, this work introduces FDBSCAN–MBKSched, a federated learning and clustering hybrid framework combining DBSCAN-based real-time abnormal health state detection and data filtering at the edge, Mini-Batch K-Means using MapReduce in the cloud, and an adaptive update scheduling mechanism. DBSCAN removes noisy data and identifies abnormal health states in real time at the edge, while non-emergency summaries are sent to the cloud for scalable clustering. The Federated Learning (FL) module governs distributed model training without sharing raw data, with devices dynamically adapting update frequencies as a function of model freshness, battery level, and event urgency. Experimental validation on real- IoMT datasets shows that FDBSCAN–MBKSched attains 12% improved anomaly detection accuracy, 21% reduced energy usage, and 17% lower emergency latency compared to traditional fuzzy clustering–based baselines. These findings demonstrate the efficiency of the framework for latency-sensitive, privacy-preserving, and resource-constrained healthcare applications.