Research On Android Malware Detection Method Based On Deep Learnin

At present,with the development of Internet technology and hardware devices,mobile device users have become the main force of the Internet,and the incidents of attacks on mobile devices have increased rapidly.Among them,mobile phone users of the Android system are the main victims of such attacks.Android malware has become one of the major threats to current cyberspace security.When traditional machine learning methods are used to detect Android malware,there are problems such as complex feature extraction process and high computational difficulty,and such methods usually only have detection results and do not explain and analyze their malicious behaviors.Therefore,this paper proposes an Android malware detection method based on deep learning to improve the detection efficiency of Android malware,and at the same time endow the Android malware detection results with interpretability to generate software malicious behavior descriptions.Traditional machine learning requires complex feature engineering in malware analysis and is not suitable for large-scale malware analysis.To improve detection efficiency on Android malware.Based on this,this paper maps Android malware bytecode files into grayscale images,and comprehensively utilizes depthwise separable convolution(DSC)and attention mechanism to propose Android malware detection based on global attention module(GCBAM).method(DSC＿At).Extract bytecode files from APK files;convert bytecode files into corresponding grayscale images;build a GCBAM-based detection model to train the image dataset to make it capable of Android malware detection.Experiments show that the DSC＿At method can effectively detect the Android malware family,and has better results in both precision and recall than machine learning algorithms.Malware detections for Android usually yield only benign or malignant results and do not reveal why the software is classified as malware.In order to reveal the malicious behavior of Android software and increase the trust in the method,an attention mechanism-based Android malware detection method(XAI＿At)is proposed.Firstly,information features are extracted from Android malware,data preprocessing is used to generate feature information,multi-layer perceptron is used for feature learning,and attention mechanism is introduced to capture sensitive features,and BP algorithm is used to train and classify the learned data.The captured features use the interpreter in XAI＿At to generate behavioral descriptions to explain the core malicious behavior of the application.Experimental results show that the XAI＿At method can effectively detect Android malware samples,and can generate behavioral descriptions based on the captured sensitive features,whose descriptions reveal the reasons for being classified as malware.In addition,the method can also explain why benign samples are falsely detected as malware.