BFSN: A Novel Method of Encrypted Traffic Classification Based on Bidirectional Flow Sequence Network
Xinxin Tong, Xiaobin Tan, Lingan Chen, Jian Yang, Quan Zheng · 2020
With the rapid development of network technology and encryption technology, network security issues have received more and more attention, and network encryption traffic is increasing, which results in a huge challenge for network traffic classification. Combining machine learning algorithms with manual design has become a mainstream approach to solve this problem, However, it requires a large amount of human effort to extract and process features, which depend on professional experience heavily. In this paper, We discuss the essential reason why convolutional neural network(CNN) can deal with the problem of encrypted traffic classification and propose a novel classification framework the Bidirectional Flow Sequence Network(BFSN) based on long short-term memory (LSTM). Compared with the traditional traffic classification scheme, the BFSN is an end-to-end classification model that learns representative features from the raw traffic and classifies them. Moreover, We apply the length and direction information of the encrypted traffic to construct the bidirectional traffic sequence and then process it based on LSTM. Our Experiments gains the excellent accuracy about 91% based the ISCX VPN-NonVPN dataset.