Detecting Malicious Server Based on Server-to-Server Realation Graph

Abstract

The rapid development of Internet attack has posed severe threats to information security. Therefore, it's of great interest to both the Internet security companies and researchers to develop novel methods which are capable of protecting users against new threats. However, the sources of these network attack varies. Existing malware detectors and intrusion detectors mostly treat the web logs separately using supervised learning algorithms. Meanwhile, using features beyond network connection content are starting to be leveraged for Internet server classification. In this paper, based on the Server-to-Server Relation Graph, we present a network Server classification method by analyzing the client distribution of each server. When constructing Server-to-Server Relation graph, k-nearest neighbors are chosen as adjacent nodes for each server node, and being compared with radial basis function network. Files are connected with edges representing the similarity of their client set. In the machine learning part, we used Label propagation algorithm, a semi-supervised learning algorithm which propagates class labels on a graph. We evaluate the effectiveness of our proposed method on a real and large dataset. Experimental results demonstrate that the precision of our method is acceptable and worthwhile.