HGDetector: A hybrid Android malware detection method using network traffic and Function call graph

Abstract

The malicious infestations of Android malware caused huge economic losses to users over the past few years. Machine learning-based malware detection enhances the accuracy and partially mitigates these security threats. However, when the static or dynamic features cannot effectively represent software behavior, the accuracy of the model will be reduced. For this issue, a multi-features hybrid malware detection and category classification method HGDetector is proposed, this approach provides a more comprehensive representation of software behavior. HGDetector first extracts the software static function call graph and constructs the network behavior function call graph, then applies the dynamic network traffic features of the software to build the node interaction graph and edge-node graph; Subsequently, these features were fused and converted into a vector representation employing graph embedding method; Finally, combined with the proposed HGDetector, different classifiers were used to test the accuracy of malware detection and category classification. The experimental results demonstrate that the fusion of hybrid features can enhance malware detection accuracy by approximately 4 % when network traffic features effectively capture APP's behavior. Conversely, in cases where network traffic features alone are insufficient to represent software's network behavior, the application of hybrid features can improve malware detection accuracy by 21 %-26 %.

Keywords

• Android
• Malware detection
• Multi-features hybrid
• Graph