The Permission Management Of Android OS Based Machine Learning

In recent years, with the popularity of smart phones continue to improve, mobile phone security issues have attracted more and more people's attention. At the same time, Google Android security strategy tends to be increasingly stringent. On the one hand, it shows that with the continuous development of Android, Google pays more and more attention to security issues; on the other hand, it also reflects the seriousness of the Android security issues from the side. In particular,there are many defects in the permissions mechanism in Android system,such as its "All-or-None" application authorization model, that is, the application can be successfully installed only if all permissions are granted. Furthermore, after the installation is completed, permissions granted could not be revoked or changed. Therefore, it is necessary to research the application of Android system software permission management technology.Aiming at the problems existing in the authority mechanism of Android system, this paper designs a kind of authority dynamic management scheme based on machine learning. The uniqueness of the program is that the use of machine learning classification technology and Android authority management combined to change the original Android authority management mechanism, which allows uesers change the dangerous permissions granted to application after the installation is complete. Thus user has a more flexible rights management that can effectively prevent malicious behavior at any time. Besides, the authority mechanism also allows users to understand as much as possible the relationship between permissions and behavior in order to minimize the risk of equipment when making permission-related decisions.The authority management scheme gives users hazard warnings and authorization recommendations when the app is running. It mainly realised through the analysis of the use of permissions in different kind of applications. There are two parts of authority management scheme in this paper: the application classification module and the authority dynamic management module. The application classification module takes the requested permission and the sensitive API, both of which extracted from the APK file, as the features of the application classification. Finally, after the training and test on the features, it outputs a classfier model. The authority dynamic management module determines the class to which the application belongs through the former classifier model, and then gives users hazard warning information based on the whitelist of the kind of application, and finally allows users to choose whether or not to grant the permissions.Based on this authority management scheme, this paper realizes a dynamic authority management system based on Xposed. The system improves and enhances the Android original authority management model,that is, to retain the original Android rights management mechanism, but when passed the original authority management, it still need additional authority management check, after which can the authority be granted to use. We use Xposed framework to HOOK the key function in system permissions check and modify its original permissions check logic, with the machine learning based on the authority management mechanism, the Android permissions check mechanism is improved.