Hybrid deep learning-enabled framework for enhancing security, data integrity, and operational performance in...

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Biomedical Engineering

Background:

• Human-centric Internet of Things (H-IoT) systems in healthcare and smart environments face critical security challenges.
• Traditional security methods are inadequate for dynamic, real-time physiological data streams, leaving H-IoT devices vulnerable.

Purpose of the Study:

• To propose a novel trust-aware hybrid framework for adaptive security in H-IoT systems.
• To enhance real-time anomaly detection and adaptive access control for physiological data integrity.

Main Methods:

• Integration of Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Variational Autoencoders (VAE) for physiological signal analysis.
• Development of a dynamic Trust-Aware Controller (TAC) for real-time trust score computation.
• Evaluation on benchmark and proprietary H-IoT datasets under various attack and noise conditions, including edge device deployment.

Main Results:

• Achieved an average F1-score of 94.3% for anomaly detection and 96.1% accuracy for access classification.
• Demonstrated 12-18% improvement in detection sensitivity compared to traditional baselines.
• Maintained real-time inference latency under 160 ms on edge hardware, showing high stability under adversarial conditions.

Conclusions:

• The proposed framework offers a scientifically grounded and scalable solution for adaptive security in H-IoT networks.
• The fusion of deep learning and trust modeling enhances responsiveness and resilience in H-IoT security.
• Paves the way for secure next-generation health and wearable ecosystems.