Tor Traffic Classification from Raw Packet Header using Convolutional Neural Network

Abstract

As the amount of network traffic is growing exponentially, traffic analysis and classification are playing a significant role for efficient resource allocation and network management. However, with emerging security technologies, this work is becoming more difficult by encrypted communication such as Tor, which is one of the most popular encryption techniques. This paper proposes an approach to classify Tor traffic using hexadecimal raw packet header and convolutional neural network model. Comparing with competitive machine learning algorithms, our approach shows a remarkable accuracy. To validate this method publicly, we use UNB-CIC Tor network traffic dataset. Based on the experiments, our approach shows 99.3% accuracy for the fractionized Tor/non-Tor traffic classification.