Abstract
The increasing reliance on Human-centric Internet of Things (H-IoT) systems in healthcare and smart environments has raised critical concerns regarding data integrity, real-time anomaly detection, and adaptive access control. Traditional security mechanisms lack dynamic adaptability to streaming multimodal physiological data, making them ineffective in safeguarding H-IoT devices against evolving threats and tampering. This paper proposes a novel trust-aware hybrid framework integrating Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) models, and Variational Autoencoders (VAE) to analyze spatial, temporal, and latent characteristics of physiological signals. A dynamic Trust-Aware Controller (TAC) is introduced to compute real-time trust scores using anomaly likelihood, context entropy, and historical behavior. Access decisions are enforced via threshold-based logic with a quarantine mechanism. The system is evaluated on benchmark datasets and proprietary H-IoT signals under diverse attack and noise scenarios. Experiments are conducted on edge devices including Raspberry Pi and Jetson Nano to assess scalability. The proposed framework achieved an average F1-score of 94.3% for anomaly detection and a 96.1% accuracy in access decision classification. Comparative results against rule-based and statistical baselines showed a 12-18% improvement in detection sensitivity. Real-time inference latency was maintained under 160 ms on edge hardware, validating feasibility for critical H-IoT deployments. Trust scores exhibited high stability under adversarial data fluctuations. This research delivers a scientifically grounded, practically scalable solution for adaptive security in H-IoT networks. Its novel fusion of deep learning and trust modeling enhances both responsiveness and resilience, paving the way for next-generation secure health and wearable ecosystems.