Officially titled: Applying Deep Learning to the classification of VPN and non-VPN encrypted network traffic
This was a project that I completed for my dissertation in my final year of university.
Due to the rapid increase in internet traffic, network traffic classification is required for Quality of Service (QoS) management. The use of Virtual Private Networks (VPNs) and encryption has made this much more difficult, as existing methods such as deep packet inspection have become futile. Hence, many researchers have begun applying Deep Learning to this problem. This project employs a payload-independent method in which traffic flows are represented as Pseudo Image Matrices (PIMs), and input into a Convolutional Neural Network to categorize them into different classes i.e., Browsing, Streaming and Email. Unlike many classifiers in current literature, only 6 packets of a flow are needed to identify its class, therefore facilitating real-time classification. To address the difficulties of classifying encrypted traffic, this project focuses on the classification of VPN and non-VPN encrypted traffic. Using the UNB ISCX VPN-nonVPN dataset, the developed classifier achieves an accuracy of 70.9%, and takes approximately 39ms to classify a flow after the arrival of the 6th packet.