EN-Pnaconv: Pnaconv-Based Intrusion Detection System for IoT
Hu Yuan, Chen Hui Min · 2024
Due to the rapid development of machine learning, neural networks are beginning to be applied to intrusion detection systems, especially those based on graph neural networks. Graph neural networks are applicable to a variety of graphical data in non-Euclidean space and can utilize a variety of features based on graphical data. Training and evaluation data for network intrusion detection systems are usually represented as network connection records, which can be represented using graph data form. In this paper, we propose a new network structure, EN-Pnaconv, which can fully utilize the edge features, node features, and topology information of network connection record graphs for network intrusion detection in IoT networks. Experimental evaluation is conducted using two latest benchmark datasets of intrusion detection systems with large differences, and the average multi-classification F1 scores of the two benchmark datasets are up to 0.81 and 0.91, respectively, which outperforms the state-of-the-art methods in terms of key classification metrics, and the results have proven that the model effectively improves the accuracy of multi classification recognition for network attacks.