Program Structure-Based Feature Selection for Android Malware Analysis

Zhou and Jiang [1] extensively surveyed and analyzed Android malware and found that 86% of the malware collected incorporated repackaged benign applications, and that many of them utilized common advertisement libraries. Such benign code reuse in malware can be expected to cause automated classification and clustering approaches to fail if they base their decisions on features relating to the reused code. To improve detection, classification, and clustering, feature selection from mobile malware must not be naive, but must instead utilize knowledge of malicious program semantics and structure. We propose an approach for selecting features of mobile malware by using knowledge of malicious program structure to heuristically identify malicious portions of applications.