Network Traffic Classification Using Semi-Supervised Approach

A semi-supervised approach for classification of network flows is analyzed and implemented. This traffic classification methodology uses only flow statistics to classify traffic. Specifically, a semi-supervised method that allows classifiers to be designed from training data consisting of only a few labeled and many unlabeled flows. The approach consists of two steps, clustering and classification. Clustering partitions the training data set into disjoint groups (¿clusters¿). After making clusters, classification is performed in which labeled data are used for assigning class labels to the clusters. A KDD Cup 1999 data set is being taken for testing this approach. It includes many kind of attack data, also includes the normal data. The testing results are then compared with SVM based classifier. The result of our approach is comparable.