Mining distinguishing patterns based on malware traces

The automatic generation of malicious behavior pattern based on system call trace is important to malware detection. This paper studied the existing generation method of malicious behavior specification. In order to reduce the complexity of pattern generation, it constructs graph which vertex label is unique, and uses these graphs to mine the pattern. To address the issue of limitation of the minimal contrast subgraph mining method, it uses multiple positive and negative samples, and proposes a mining method to mine distinguishing patterns based on mutual information. It designs the overall framework of mining process, and gives the mining algorithm. Finally, validation results demonstrate the effectiveness.