Feature selection for optimizing traffic classification
Computer Communications
| Hi-index | 0.00 |
New application traffic occurring on Internet frequently challenges the traditional traffic classifiers based on machine learning. These classifiers always identify it inaccurately and assign it into one of their known classes forcibly, even though the extra class is labeled as 'other' when training. In this case, the precision of identifying known classes is reduced. In this paper, a robust traffic classification framework based on OC-SVM combined with MC-SVM (TCFOM) is presented. We capture several kinds of application traffic, and carry out an experiment under supervised environment. Using the OC-SVM, the unknown traffic is classified into extra class labeled as 'other'. The precision of identifying known traffic is improved. Using the unknown traffic identified, the new classifying model is set up. TCFOM can classify the unknown traffic and extend well. We compare TCFOM with three classifiers respectively based on SVM, RBF network, Naive Bayes. Experimental results show that the robustness of TCFOM is best.