Machine Learning Based Android Malware Detection using Significant Permission Identification

The increasing popularity of smartphones andtablets has introduced Android malware which israpidly becoming a potential threat to users. A recentreport indicates the alarming growth rate of Androidmalware in which a new malware is introduced inevery second more precisely in 10 seconds. To againstthis dangerous malware growth, this paper proposesa scalable malware detection system using permissionanalysis behavior that can identify malware appseffectively and efficiently. We propose multi-level ofpruning procedures to identify the most significantpermission instead of extracting all permissions. Thepropose system utilizes supervised classificationmethod in machine-learning to classify differentfamilies of benign and malware apps. We found that22 permissions are significant actually. Ourevaluation finds that the analysis time of using these22 permissions are 4 to 32 times less than using allpermissions. The results show that most of malwareapps are located the unnecessary permission onAndroidManifest.xml to inject the malicious codes inthe apps