Image-based Encrypted Traffic Classification with Convolution Neural Networks

Abstract

Network traffic classification plays an important part in the network management and network monitoring. It can help administrators to understand the constitution of network traffic, facilitate administrators to manage the network and provide differentiated service quality and security monitoring. However, the widespread usage of the encryption techniques and dynamic ports policy make encrypted traffic classification become a great challenge for traditional traffic classification methods. In this paper, we propose an image-based method that can classify encrypted network traffic with a high accuracy. The basic idea of the method is to convert the first few nonzero payload sizes of session to gray images, and classify the converted gray images with convolutional neural network to achieve the goal of categorizing the encrypted network traffic. This method is very light-weight and it can automatically extract features, select features and classify encrypted network traffic to categories. We use the public dataset ISCX VPN-nonVPN to validate our proposed method. The experimental results show that our proposed approach achieves F1 score of 97.73% on the conventional encrypted traffic classification and F1 score of 99.55% on the virtual private network traffic(VPN) classification.