Investigation of machine learning based network traffic classification

Timely and accurate traffic classification and application characterization are becoming increasingly important with many applications in wired and wireless networks, e.g., traffic engineering, security monitoring, and quality of service (QoS). In particular, Software Defined Networking (SDN) is a new networking paradigm that has great impact on future IP networks and 5G wireless networks. In SDN networks, application awareness is essential for functionalities such as virtual network resource slicing and fast routing. Compared to traditional classification methods such as port-based and payload-based algorithms, machine learning (ML) approaches offer a better choice in Internet traffic characterization by using payload-independent traffic statistics. In this paper, two ML algorithms, namely supervised Support Vector Machine (SVM) and unsupervised K-means clustering, are studied for traffic classification. It has been found that an overall accuracy of over 95% can be achieved. Meanwhile, the system performance can be further improved with model tuning and feature selection.