Efficient Malicious Traffic Classification Methods based on Semi-supervised Learning

Abstract

The proliferation of mobile communication systems, arrival of high-speed broadband networks and more complex network topologies have exacerbated cyber-threats. Cyber-warfare has become an aspect of modern war-fare that can no longer be overlooked. In recent years, network intrusions launched using the Internet have seriously undermined the security systems of many nations. Classifying malicious network traffic is the first step in network intrusion detection. In this paper, we propose three models using semi-supervised learning-based malicious traffic classification (MTC) methods that effectively improve the classification of traffic using a small proportion of labeled traffic data. Employing three different deep neural networks as feature extraction networks respectively, the proposed models use transductive transfer learning and domain adaptive ideas, and ladder networks as classification layers. Experimental results are provided to validate the proposed methods.