A Multi-level Internet Traffic Classifier Using Deep Learning

In the network domain, there is a continuous ongoing evolution in the type of connected devices and the nature of developed applications. This presents the network with varying Quality of Service (QoS) and security requirements. Consequently, there is a need to classify Internet traffic to facilitate its management. Thus, a granular classification based on needs is required, one which relates better to the network requirements of the traffic. In this paper, we apply deep learning to classify traffic requiring different QoS and security policies. We propose a multi-level classification framework applying a new data representation method. A comparison between the proposed data representation method and a previous method is presented. The implementation results show that the proposed data representation model outperforms the previous one and promises to permit the classification of the traffic at different granularity. Recording up to 95% accuracy using only the size, interarrival time, direction, transport protocol of the first 16 packets of each flow, our method can be employed in an online classification platform.