A novel semi-supervised learning method for Internet application identification.

Several methods based on port, payload, and transport layer features have been proposed to detect, identify, and manage Internet traffic. The diminished effectiveness of port-based identification and overheads of deep packet inspection methods motivated us to identify Internet traffic by combining distinctive flow characteristics with the machine learning method. However, the abundant ground truth Internet traffic, which is important for building a supervised classifier, is difficult to be obtained in real conditions. In this study, we propose a semi-supervised learning method that combines further division of recognition space technique with data gravitation theory. The further division of recognition space classifier is a powerful multi-classification tool that can be helpful for multi-application identification. The data gravitation may reveal the underlying data space structure from unlabeled data, and thus, it is integrated into the classification to develop a better classifier. The experimental results on the real Internet application traffic datasets demonstrate the advantages of our proposed work. [ABSTRACT FROM AUTHOR]