AI-Driven Hybrid Detection and Classification Framework for Secure Sleep Health IoT Networks

Background: Sleep disorders like insomnia, obstructive sleep apnea (OSA), REM behavior disorder etc. are nowadays diagnosed through the Internet of Things (IoT)-enabled sys-tems that monitor and analyze the subject's sleep data. Health IoT networks are rife with communications of sensitive physiological data from wearable EEG, ECG, SpO₂ and res-piratory sensors. However, these networks face threats from anomalous traffic flows, sig-nal sabotage and data integrity violation. In this paper, an AI-based hybrid detection and classification framework is proposed for secure Sleep Health IoT (S-HIoT) networks. The integrated CNN, BiLSTM and RF model provides a proposed framework for joint sleep-stage classification and network anomaly detection. To this end, a multi-objective loss function is proposed for jointly optimizing the physiological state prediction and se-cure traffic monitoring. Experimental validation using the Sleep-EDF and CICIoMT2024 datasets demonstrate a classification accuracy of 97.8% for sleep staging, and 98.6% for network detection with low inference latency (