Multimodal Neural Network Based Malware Detection for Android
Fuxuan Gu, Zhibo Du · 2024
Malware poses a widespread and serious threat to mobile devices, and existing machine learning-based detection methods have problems such as the need for complex feature engineering and incomplete detection types when facing large-scale and high-precision malware detection. Based on this, a method for Android malware detection based on the combination of multimodal neural networks and static analysis methods is proposed: pseudo-dynamic and static program analyzers are used to extract Android application-related features, including permissions, opcodes, and API call sequences, respectively. Malware detection is realized by fusing and classifying different types of features through multimodal neural networks. This approach enriches the data types compared to traditional methods and can capture and utilize the information of Android programs more effectively. Experimental results on the MalMem experimental dataset show that the proposed method has a higher comprehensive detection capability and can achieve an accuracy of 98.9% compared to existing detection methods.