A Deep Feature Fusion Method for Android Malware Detection

Abstract

In recent years, there is a rapid increase in the number of Android based malware. To protect users from malware attacks, different malware detection methods are proposed. In this paper, a novel static method is proposed to detect malware. We use the static analysis technique to analyze the Android applications and obtain their static behaviors. Two kinds of behaviors are extracted to represent malware. One kind of behaviors is the function call graph and the other kind is opcode sequences. To automatically learn behavioral features, we convert the function call graphs and opcode sequences into two dimensional data, and use deep learning method to build malware classifier. To further improve the performance of the malware classifier, a deep feature fusion model is proposed, which can combine different behavioral features for malware classification. The experimental results show the deep learning method is effective to detect malware and the proposed fusion model outperforms the single behavioral model.