FedEdgeHealth: an uncertainty-aware federated edge intelligence framework for real-time multi-modal health monitoring

Abstract This study introduces a sustainability-focused, uncertainty-aware federated edge intelligence framework for multimodal healthcare monitoring within simulated AIoT-edge environments. Unlike conventional healthcare AI systems that optimize predictive accuracy or latency in isolation, the proposed framework integrates probabilistic multi-task learning with adaptive federated coordination to enable selective client participation and risk-aware resource management across distributed edge nodes. The framework captures predictive uncertainty while learning from heterogeneous, multimodal healthcare data distributed across non-IID clients. Furthermore, an energy-aware workload management strategy is incorporated to minimize computational and communication overhead, thereby supporting sustainable edge-cloud healthcare infrastructures. Experimental evaluations using diverse public healthcare datasets in a simulated federated edge environment demonstrate that the proposed framework achieves superior predictive performance (AUC-ROC = 0.956), enhanced calibration reliability, and reduced estimated latency and modeled energy consumption compared to centralized and conventional federated learning baselines. These findings highlight the effectiveness, robustness, and resource efficiency of the proposed approach under controlled simulation conditions, establishing a reproducible methodological foundation for future physical hardware deployment and prospective clinical assessment.