Attack Traffic Detection Based on LetNet-5 and GRU Hierarchical Deep Neural Network

The paper converts the network traffic information about a single-channel grayscale image as input data. In addition, a deep hierarchical network model is designed, which combines LetNet-5 and GRU neural networks to analyze traffic data from both time and space dimensions. At the same time, two networks can be trained simultaneously to achieve better classification results because of the reasonable network association method. This paper uses the CICID2017 dataset, which contains multiple types of attacks and is time-sensitive. The experimental results show that, through the combination of deep neural networks, the model can classify attack traffic with extremely high accuracy.