An empirical study of morphing on behavior-based network traffic classification

摘要: With the rapid advancement of traffic classification techniques, a countermeasure against them called network morphing, which aims at masking to degrade performance identification and classification, has emerged. Although several morphing strategies have been proposed as promising approaches, very few works, however, investigated their impact on actual performance. This work sets out fulfill this gap from an empirical study point view. It takes into account different exerted packet size PS and/or inter-arrival time IAT evaluates by simulation. The is evaluated using three popularity used algorithms, including C4.5, Support Vector Machines , Naive Bayes, with various metrics considered. results show that not all can effectively thwart classification. Different perform distinctively in degrading identification, among integration morphings best, PS-based method alone worst. Furthermore, classifiers also exhibit distinct robustness C4.5 being most robust Bayes weakest. Finally, our shows learn nontrivial information merely direction patterns, partially explains weak protection methods because they fail take patterns consideration. Copyright © 2013 John Wiley & Sons, Ltd.