Home // ICIMP 2012, The Seventh International Conference on Internet Monitoring and Protection // View article

Parameter Estimation for Heuristic-based Internet Traffic Classification

Authors:
Michael Finsterbusch
Chris Richter
Jean-Alexander Müller

Keywords: flow classification, Internet traffic, traffic identification

Abstract:
Accurate traffic classification is necessary for many administrative networking tasks like security monitoring, providing Quality of Service and network design or planning. We apply 18 machine learning algorithms to classify network traffic based on six classes of statistical parameters. In contrast to other studies, we use a per-packet approach instead of a per-flow approach to make it possible to use the classification results for real-time network interception. In this paper we illustrate the accuracy of the algorithms with different parameter combinations with the goal to reduce the amount of necessary parameters needed for high accuracy traffic classification. Our results indicate that some parameter combinations can be used to classify a large number of protocols. We identified algorithms with good and worse classification accuracy and algorithms which need much time for classification, so that they cannot be used for real-time classification.

Pages: 13 to 22

Copyright: Copyright (c) IARIA, 2012

Publication date: May 27, 2012

Published in: conference

ISSN: 2308-3980

ISBN: 978-1-61208-201-1

Location: Stuttgart, Germany

Dates: from May 27, 2012 to June 1, 2012