Target: category-based android malware detection revisited

Jiaping Lin, Xingwen Zhao, Hui Li

Abstract: Smartphones are becoming increasingly popular in daily routines around the world. However, malware in smartphones is getting more prevalent, and will introduce potential risks to smartphone users. In this paper, we propose a new system, called Target, for detecting malware in the Android operating system, featuring both static and dynamic analysis. Our static analysis is based on user permissions, signatures and source code, and our dynamic analysis is based on the behavior of running mobile applications. A highlight of Target is its ability to reduce the probability of false positives based on the category of applications. Target first generates risk values of the mobile application being analyzed, indicating the degree of risks involved. It then uses a machine learning algorithm, named OKNN, to determine which class an application belongs to. Compared to previous work, Target is able to achieve a significant improvement in terms of malware detection accuracy.