A New Malware Classification Approach based on Statistical Feature

Abstract

Malware variants which use obfuscation and metamorphism techniques are seen as a critical security threat. Machine learning based malware classification techniques are able to discriminate different malware families and improve existing anti-malware tools. However, high dimensional feature space brings a higher time overhead and one-sided feature can decreases the accuracy. To solve this issue, we propose a statistical feature based malware classification approach and a new feature selection method which can select strong discriminative features. The results demonstrate that the proposed approach can classify modern malware variants effectively.