EDW-voting: Robust realtime traffic classification combined with flow side information

Network traffic classification has played a vital important role in network management, from security monitoring to quality of service control. Different from deep packet inspection (DPI) and port-based methods, machine learning (ML) based methods can recognize high level applications through flow statistical features other than application-level contents or fixed port numbers. In this paper, we propose the exponentially decaying window voting (EDW-Voting) method, aiming to improve the ML classification results through exploiting side information between the flows to be classified. EDW-Voting is not only effective in promoting the classification process but efficient for time consuming and memory usage. Traffic classification evaluations have been conducted on a real-word network trace. The results of average overall accuracy and per-class F-measure demonstrate that EDW-Voting effectively improves the ML classification results and outperforms fixed size sliding window voting (FSW-Voting) in all experiments.