Android Privacy Protection Based On Ensemble Machine Learning

Along with the rapid development of the mobile Internet, the number of mobile intelligent terminals which equipped with Android system grows rapidly, mobile intelligent terminals which equipped with Android system has became an indispensable auxiliary tool in people’s daily life. However, as people growing in depending on mobile intelligent terminals, serious security problems exposed. One of that is damaging the interests of users by stealing users’ private data. This thesis analyzes existing research programme on Android security, and based on this, proposes a detecting program based on ensemble classification machine learning algorithm. With this detecting program, malicious applications which would steal users’ private data can be detected, and people’s privacy could be protected.The main work is as follows:1. Rsesearch on Android security is mainly from two aspects: prevention and detection. Most research on prevention side get a real complex security model and the Android system need to be rebuilt and this make this kind of researchs full of unserviceability. Research on detection side gets its own problems too. Traditional static detection method is only valid for known malicious applications and features and unvalid for unknown malicious code and features. Dynamic detection technology tends to affect normal operation of the system becauseof its resource-highly-taken-up in system. Based on the problems above, this thesis takes a reaserch and wants to find a general detecting method which also could integrate advantages of both static detection and dynamic detection.2. With the effectiveness of prediction of machine learning classification algorithm on classification problems, this thesis takes another reaserch and wants to find whether machine learning classification algorithm could be used on Android detection or not.The main results are as follows:1. Based on reaserch on security mechanism and privacy threats on Android, this thesis finds the importance of Android permissions for user privacy theft and the reasonableness of using Android permissions in machine learning algorithm as features.2. Based on reaserch on classic machine learning classification algorithm and popular ensemble learning algorithm, this thesis designs several ensemble learning algorithm using AdaBoost and bagging method with native bayes and decision tree as basic classification algorithm, and with learning and testing, this thesis find one best ensemble learning algorithm from them.3. Base on ensemble learning algorithm above, this thesis proposes a detecting program which has advantages of both static detection and dynamic detection. And through experiments, this thesis verifies the effectiveness of this programme. In the end, this thesis compares this program with programmes proposed in current research.