Multi-Phase Traffic Classification Based on Payload

Abstract

While the internet is gaining more and more importance in our daily life, the number of applications used via the internet are increasing at the same speed. Today, fast and accurate classification of data packets transmitted over the network based on the applications has become an important issue in terms of security as well as network management. In this study, with the proposed classification approach, it is aimed to determine which application these network packets belong to, by inspecting their payloads. To classify packets, a multi-phase method based on majority voting is proposed. This method is based on training deep learning-based classifiers using different numbers of packets and updating the classification prediction as the number of packets in the network flow increases. This updated prediction is achieved by majority voting by using the predictions of previous classifiers trained by smaller number of packets from flows. With this approach, more accurate classifications can be made with less number of packages and this allows an early classification without waiting for more packages to arrive. This approach has been tested on real data collected for various applications.