Secure IoTNet: a graph-residual adversarial network integrated with Hawk-Bee optimizer for intrusion detection in IoT wireless networks

Secure IoTNet: a graph-residual adversarial network integrated with Hawk-Bee optimizer for intrusion detection in IoT wireless networks

연구 분야: Networking

학회: International Journal of Data Science and Analytics

Intrusion detection and prevention are, therefore, two of the very critical elements involved in the process of securing an IoT-based wireless communication network. The greater the number of IoT devices that proliferate, the greater the threats of advanced cyber-attacks against the sensitive data and system integrity of these networks. Traditional IDS solutions usually find the modern wireless environment difficult to handle because of its dynamic nature and complexity, thus opening a huge gap for advanced solutions. This paper responds to these challenges with the proposal of a new deep learning-based intrusion prevention framework, by combining state-of-the-art models to improve the accuracy of detection and reduce network vulnerabilities. The proposed method uses the graph-residual adversarial network (GRANet) for an effective intrusion detection in IoT-enabled wireless communication networks. The spatial and temporal patterns of network traffic have been captured in an attempt to provide an anomaly detection mechanism that performs on a plethora of constantly evolving attack vectors. Also, a hybrid optimization mechanism, Hawk-Bee Stride Finder (HBSF) is developed to fine-tune the convolution stride parameters to better extract the features by the proposed model. The novelty of this work lies in the synergistic integration of adversarial learning, residual connections, and graph-based modeling, further supported by swarm intelligence-driven hyperparameter tuning. The results, supported by extensive experiments using benchmark IoT intrusion datasets, have shown that the proposed framework outperforms the state-of-the-art models existing in terms of precision, recall, and general detection rates.