An Abnormal Traffic Detection Based on Attention-Guided Bidirectional GRU
Hao Li, Erlu He, Chunxu Kuang, Xiaopeng Yang, Xiangbo Wu, Zhe Jia · 2022
With the development of network technology and the diversification of network applications, network environment is becoming increasingly complex which leads to various network attacks. Malicious traffic, such as DDOS and infiltration attacks, may degrade the network quality, or even block the network directly. Now, traditional network security methods are difficult to meet the needs for network security and information security. Due to the high recognition accuracy and strong generalization of deep neural network (DNN), deep learning based abnormal traffic detection methods have become a research hotspot. However, there are obvious problems containing high false positive rate poor robustness. We propose a novel abnormal traffic detection method based on attention-guided bidirectional gated recurrent unit (Att-BiGRU). Different from the traditional feature extraction methods in DNNs, we use stacked sparse Auto-encoder (SSAE) to extract traffic features for interference factors filtering. The extracted features can reproduce the original input samples as much as possible which enhances the representation of traffic features and effectively improves the robustness of anomaly detection traffic. Then, attention mechanism was introduced to optimize the traffic features. Extensive experiments on UNSW-NB15 show that the proposed scheme has the impressive performance. Compared with recent abnormal traffic detection algorithms, the proposed scheme achieves better results with 89.67% accuracy and 4.2% FAR.