Reliable healthcare IoT scheduling using calibration and uncertainty aware learning

Healthcare Internet of Things (IoT) environments generate highly dynamic and heterogeneous communication traffic that requires reliable and low-latency scheduling under evolving healthcare-oriented communication conditions. Conventional scheduling approaches primarily emphasize predictive accuracy while insufficiently addressing calibration stability and robustness under temporal distribution shifts, potentially leading to unreliable scheduling decisions during deployment. To address these limitations, this paper proposes META-TASC (Meta-Learning based Temporal-Aware Scheduling and Calibration), a meta-learning-based scheduling framework for uncertainty-aware healthcare IoT systems. The proposed framework integrates heterogeneous baseline learners with meta-feature construction, uncertainty estimation, adaptive confidence fusion, and calibration-sensitive optimization to improve scheduling reliability under non-IID temporal conditions. META-TASC constructs uncertainty-aware scheduling representations using aggregated prediction probabilities, entropy-based confidence estimation, and inter-model disagreement analysis across multiple temporal observation windows. Extensive experiments were conducted using communication-layer IoT telemetry streams under 1 s, 3 s, 5 s, and 10 s observation windows, with additional healthcare-specific IoMT validation used to examine domain transferability. Experimental results demonstrate that the proposed framework maintains stable predictive behavior and improved calibration performance under dynamically varying communication conditions. Comparative evaluation further indicates that META-TASC achieves competitive predictive capability while providing improved confidence reliability and temporal robustness across evolving IoT communication conditions relevant to healthcare-oriented deployments.