Semi-supervised Encrypted Traffic Classification Using Composite Features Set

Many network management tasks such as managing bandwidth budget and ensuring quality of service objectives rely on accurate classification of network traffic. But the statistical features of encrypted traffics are not stable and do not contain sufficient information for classification all the time. Some applications support multiple protocols, and the behaviors of these applications are complicated and can’t be classified utilized only statistical features accurately. Regarding this, we propose composite features-based semi-supervised encrypted traffic classification. This is the first step utilizing composite feature set for classifying encrypted traffic. And the proposed approach is semi-supervised, fast and accurate classifiers can be obtained by training with a small number of labeled flows mixed with large number of unlabeled flows. We conduct the experiments to evaluate the performance of the proposed approach, obtaining promising results. Keywords-Traffic classification; encrypted traffic; composite feature set; semi-supervised learning