- 期刊
A Feature-Based Network Traffic Classification Approach
摘要
A novel traffic classification approach using deep learning was proposed in this paper. Most current deep-learning-based traffic classification models use raw traffic data as the input of the neural networks. However, the amount of raw data is overwhelming. Many current models use a subset of the raw traffic data, such as the first hundred bytes of a network traffic flow. This idea limits the size of neural network input but loses some traffic information that helps classify traffic types. Instead of using raw traffic data, in this paper, we use traffic flow features as the neural network input. To efficiently classify traffic types, we analyzed what features play important roles in traffic classification and designed neural networks accordingly. Experimental results show that our approach increases classification precision by 19 percent compared to state-of-the-art methods. Meanwhile, if trained by a dataset labeled normal traffic and intrusion traffic, our models can also be applied to network intrusion detection and cybersecurity.