Toward Accurate Efficient Online Traffic Classification First Year Report

Wei Li
2007-01-01
Abstract:This report documents my work in the first year as a research st uden . I devised a classification method for accurate online network ap plication classification. It does not rely on port numbers, works without inspect ion of packet payload but collects statistical features, using information from the headers of first several packets. I used a semi-automatic supervised machine learni ng approach to build the feature set and to tune the parameters, in order to train t he final classifier. I codesigned a prototype feature-capture and classification in frastructure and carried out experiments to evaluate the method based on two real pack et traces. Finally, only 12 features are collected from 5 packets at the start of e ach connection flow. The resulting classifier recognised different types of appl ications with up to 99.8% total accuracy.
What problem does this paper attempt to address?