Methods to Select Features for Android Malware Detection Based on the Protection Level Analysis

Abstract

Android’s permission system is asked to users before installing applications. It is intended to warn users about the risks of the app installation and gives users opportunities to review the application’s permission requests and uninstall it if they find it threatening. However, not all android permissions ask for the user’s decision. Those who are defined as ‘Dangerous’ in the permission protection level are only being confirmed by the users in Android Google Market. We examine whether the ‘Dangerous permissions’ are actually being a main component of detection when it comes to defining the app as malicious or benign. To collect important features and to investigate the correlation between the malicious app and the permission’s protection level, feature selection and deep learning algorithms were used. The study evaluates the feature by using the confusion matrix. We used 10,818 numbers of malicious and benign applications, and 457 permission lists to investigate our examination, and it appeared that ‘Dangerous’ permissions may not be the only important factor, and we suggest a different perspective of viewing permissions.