Machine learning methods for Android malware detection

Abstract

With the Android mobile device becoming increasingly popular, the Android application market has become a main target of the malware attacks. Therefore, many methods have been used to protect the mobile application users from being attacked. However, those methods have shortcomings in detecting the malware within a short time, and can be easily bypassed. To detect the malware before the installed time, and overcome the drawbacks of dynamic analysis and signature based analysis, the machine learning based malware detection methods has been proposed. In this project, I have adopted this approach to develop a tool to extract Android application features, and built the classification model using the generated feature sets. The result shows that classification the model can reach 98% accuracy in predicting the maliciousness of the application. I have also generated the transformation attack samples, which will be used in further machine learning based malware detection studies.