ME-SHMS: Adaptive Unsupervised Anomaly Detection for Resilient Edge-based Remote Patient Monitoring

This is the authors' abstract. We don't add key points for this paper.

In this paper we propose MESHMS (Smart Health Monitoring System), an edge assisted RPM architecture for intermittent connectivity and scarce labeled clinical anomaly data. The four-layer architecture processes patient telemetry in real time; FIFO flash buffers at the edge gateway preserve temporal ordering during network disruptions; unsupervised conservative OR fusion combines DAE with IF without need for clinical anomaly labels for model fitting or threshold calibration. An EWMA threshold dynamically adapts the DAE decision boundary to gradual physiological baseline drift. Evaluated on the MHEALTH and synthetic multisensor datasets, the OR fused DAE and IF configuration achieved 96.71% accuracy and an F1 score of 0.9317. Deployed on an ARM Cortex A72 edge resource profile, the pipeline required 38.5 MB of RAM with an inference latency of 3.8 ms. Across five simulated network outages ranging from 5 minutes to 24 hours, zero packet loss occurred while maintaining full FIFO sequence integrity. Under an 8% simulated baseline drift, the EWMA threshold with α = 0.01 reduced false positives by 64.5% compared to a static threshold. While these findings demonstrate technical feasibility under benchmark and simulated outage conditions, clinical readiness still requires prospective field and clinical validation.