Collaboratively filtering malware infections: A Tensor decomposition approach

Malicious applications pose a threat to the security of the Android platform and Android smartphones often remain unprotected from novel malware. In this paper, we propose a Tensor filter, a collaborative approach for detection of Android malware, which extracting three feature sets into a three-order tensor A instead of a vector space. As the limited resources impede monitoring applications at run-time, Tensor filter performs a broad static analysis, gathering three contributed feature sets by a mathematical statistics method. These features are coded into a three-order tensor A and fitted an integrated tensor A^∗ as well as deal with sparse problem by Tensor decomposition, which could reveal latent factors together. Tensor filter divides a large scale unknown applications into two categories, benign or malicious, according to integrated tensor A^∗, and typical combination of features indicative for malware can be used for explaining the decisions of our method. In an evaluation with 60,420 applications and 10,000 malware samples Tensor filter outperforms several industrial malware detection tools with the accuracy of 82.5%.