Machine Learning Based Android Malware Detection using Significant Permission Identification

Abstract

The increasing popularity of smartphones and tablets has introduced Android malware which is rapidly becoming a potential threat to users. A recent report indicates the alarming growth rate of Android malware in which a new malware is introduced in every second more precisely in 10 seconds. To against this dangerous malware growth, this paper proposes a scalable malware detection system using permission analysis behavior that can identify malware apps effectively and efficiently. We propose multi-level of pruning procedures to identify the most significant permission instead of extracting all permissions. The propose system utilizes supervised classification method in machine-learning to classify different families of benign and malware apps. We found that 22 permissions are significant actually. Our evaluation finds that the analysis time of using these 22 permissions are 4 to 32 times less than using all permissions. The results show that most of malware apps are located the unnecessary permission on AndroidManifest.xml to inject the malicious codes in the apps.