Android Malware Application Detection Method Based on RGB Image Features in E-Commerce

Abstract

In recent years, with the continuous development of Android applications, part of e-commerce business gradually promoted to the Android platform. Android malicious applications have become an important factor threatening E-Commerce security. The Android malware application detection methods based on machine learning play an important role in malware detection, and most of them construct self-defined structured features with static analysis technique. The accuracy and comprehensiveness of features are disturbed by shell and code obfuscation techniques. In the case of applications using shell and code obfuscation techniques, these techniques would lead to instability of detection result. In addition, the way of these methods, which mainly use disassembly technology to extract source data, would deteriorate the efficiency of detection. In view of such problems, we propose an android malware application detection method based on RGB image features. Our method uses RGB image visualization technology to directly transform binary files into unstructured RGB images. In the process of model training, we acquire advanced features autonomously by training the VGG16 model with RGB images. We perform a comprehensive analysis of our approach and other methods on the Android malware dataset. The results show good efficiency and the adaptability of our method.