TCP Traffic Classification Using Relaxed Constraints Support Vector Machines

Abstract

The traffic classification problem is critical for management, security monitoring, and traffic engineering in computer networks. It has recently taken into consideration by both network operators and researchers. It allows network operators to predict future traffics and detect anomalous behavior and also allows researchers to create traffic models. In this paper, we use a new architecture of support vector machines, namely relaxed constraints support vector machines (RSVMs), to present a traffic classifier that can achieve a high accuracy without any source or destination address or port information. We just use packet length to predict the application class for each flow. RSVM is an efficient and noise-aware implementation of support vector machines that assigns an importance degree to each training sample in such a manner that noisy samples and outliers are given a less degree of importance. Experimental results with UNIBS and AUCKLAND, two sets of traffic traces coming from different topological points in the Internet, show that the proposed classifier is more reliable and has better accuracy.