TrustAware-GNN: Graph-Neural-Network-Based Trust Management for IoT Anomaly Detection
Kamran Ahmad Awan, Ikram Ud Din, Ahmad Almogren, Zhu Han, Mohsen Guizani
Abstract
The widespread deployment of Internet of Things (IoT) devices has intensified the demand for scalable and secure trust management IoT systems. Existing GNN-based approaches often neglect real-time adaptability and contextual trust in dynamic, heterogeneous networks. This study introduces TrustAware-GNN, a trust-aware graph neural network framework designed to robustly evaluate device trustworthiness in IoT environments. The model integrates a multi-dimensional trust mechanism encompassing direct, indirect, temporal, and contextual trust, computed using localized device parameters: reliability, capability, security posture, reputation, and location awareness. Trust values modulate edge weights within the graph, enabling trust-adaptive message propagation. A trust-threshold-based edge formation mechanism filters unreliable links, while attention-based aggregation refines node embeddings. The model continuously adapts to behavioral shifts via time-decayed trust updates and contextual similarity matching. Simulation was conducted across IoT-23, EDGE-IIoTSET, AutoTrust, and ToN-IoT datasets. TrustAware-GNN achieved 94.83% accuracy on EDGE-IIoTSET and 93.25% on IoT-23, outperforming MGNN, STAR-GCN, and SEGC-PP in both accuracy and adaptability under dynamic trust scenarios.