Aiming at the current problems of imbalanced network flow data and over-reliance on labels in the field of intrusion detection for the Internet of Things,an intrusion detection method based on a self-supervised graph attention network is proposed.By employing a masked autoencoder,the model's learning representation capability for graph-structured data is enhanced,extracting useful information from unlabeled data.Additionally,the graph attention network is improved adaptively learning features from nodes and edges,allowing the model to better extract sample features.Finally,a multi-layer perceptron is utilized for the multi-class classification task of attack samples.Experimental results on the UNSW-NB15 and ToN-IoT datasets demonstrate that the average F1 scores across all categories achieved by this method are 99.04%and 99.76%,respectively,indicating its effectiveness in completing intrusion detection tasks with high robustness.
维普
万方数据