Abstract

As IoT ecosystems grow increasingly complex and dynamic, traditional anomaly detection systems fall short often overlooking the subtle, context-driven deviations that precede critical security breaches. This seminar introduces a context-aware, graph-based anomaly detection framework designed to meet the demands of modern IoT . The proposed method models network traffic as a time-evolving multi-edge graph, capturing not only communication flows but also contextual and knowledge-based interactions between devices. By applying scalable community detection, the system identifies stable interaction patterns, enabling the detection of structural, temporal, and behavioral anomalies. These insights are integrated into a heterogeneous Graph Neural Network (HeteroGNN) that classifies network edges with high precision, enabling robust, real-time detection of both known and emerging threats. Experimental evaluations on benchmark datasets including CIC-ToN-IoT and CIC-IDS2017 demonstrate the framework’s effectiveness in accuracy, adaptability, and speed, confirming its suitability for deployment in real-world, resource-constrained environments.

Addintional Information

You can take part in the LRE Open Space either online or in person. For participants attending in person, please register via https://forms.office.com/e/eQrevbmbUh before 2024-04-22, for organisational reasons.