A novel detection model for abnormal network traffic based on bidirectional temporal convolutional network

The increasingly complex and diverse network environment has increased traffic intrusion behaviors, but the traditional machine learning-based model has the problems of time-consuming and low detection accuracy due to the need of manually selecting features. Therefore, it is very important to construct an automatically abnormal network traffic detection model with a high detection accuracy. The goal of this paper is to train the network traffic through deep learning technology to generate an automatic abnormal network traffic detection model without manual design of features. We propose an abnormal network traffic detection model called BiTCN based on bidirectional time convolution network, it first uses temporal convolutional network (TCN) model to better grasp the sequence characteristics of network traffic, and then uses Exponential Linear Unit (ELU) activation function to replace ReLU in the model training stage to avoid the problem of neuron “death” leading to the reduction of detection accuracy, as well as improves the original one-way model to a two-way model to capture the two-way semantic fusion characteristics of network traffic. We evaluate the efficiency and effectiveness of the proposed BiTCN model by comparing it with different models on the CTU and USTC-TFC2016 datasets. The experimental results show that the proposed BiTCN model outperforms other models in terms of the precision, accuracy, recall and F1-measure. In this paper, we propose a novel detection model for abnormal network traffic based on bidirectional temporal convolutional network , it solves some shortcomings and limitations of existing models, and obtains a high detection accuracy of abnormal network traffic with a high stability.