High accurate internet traffic classification based on co-training semi-supervised clustering

Currently the popular methods of network traffic classification are the classification based on payload and supervised or unsupervised machine learning algorithm. But in the actual flows classification, traditional methods have faced more and more challenges due to increasing applications and difficult to obtain labeled flows. This paper proposes a traffic classification method based on co-training semi-supervised clustering. This method uses a few labeled flows and classifiers based on two different evaluation metrics to achieve high-performance classifiers. Finally we intercept data from the campus backbone and use open source tools to implement the experiment, which shows higher accuracy, precision and recall than other classic clustering methods (such as K-means, DBSCAN and two-layer semi-supervised clustering).