MSNdroid
Xiaoxia Qin, Fangping Zeng, Yu Zhang · Proceedings of the ACM Turing Celebration Conference - China · 2019
Android operating system has become a very popular mobile operating platform. However, the popularity and openness of the Android has also made it a major target for malicious application developers. In recent years, many researchers have conducted research on Android malware detection, but almost all static analysis techniques focus on the analysis of Manifest.xml and java-layer code. This results many malware producers hide malicious code in the native-layer to evade existing detection techniques. Therefore, in this paper we innovatively add the feature of native-layer to the features data. We extract the corresponding native_apis feature, combined with permissions feature and system_apis feature to form complete features. Using the Deep Belief Networks (DBN) algorithm, we achieve classification accuracy of 98.71% and false negative rate of 0.7%. To the best of our knowledge, this is the first study and MSNdroid is the first tool to apply deep learning to native code features for Android malware detection.