Research And Application Of Static Detection Method Of Android Malicious Code Based On Machine Learning

The security detection of Android applications has become one of the hot research issues in the field of network security.In order to improve the accuracy of Android malicious code detection,based on machine learning methods,this article conducts research on static detection methods for Android malicious code from two perspectives: detection after converting malicious code files into grayscale images and detection based on semantic analysis.A machine learning based Android malicious code detection system is designed and implemented for practical application scenarios.The main research work and achievements include:1.Proposed an Android malicious code graph detection method based on attention mechanism.First,by parsing the apk file,binary Android malicious code files are converted into grayscale images.Then,by using CNN as the basic framework and Res Net as the backbone network,self attention mechanism is introduced to design an Android malicious code graph detection model based on attention mechanism.Among them,the self attention mechanism learns the activation of specific samples through a channel dependent self selection gate mechanism,enabling them to use global information and selectively emphasize information features,suppressing unimportant feature information.The experimental results in environments such as CNN convolutional neural network Lenet5,residual neural network Res Net,and VGG16 show that the malicious code graph detection model based on self attention mechanism performs best.2.A semantic analysis based Android malicious code detection method has been proposed.This method identifies malicious code on Android devices through in-depth analysis and learning of Semantic information in terms of code structure,variable use,function call,etc.of malicious applications,and training through semantic analysis based models.The experimental results in environments such as DNN neural network,LSTM recurrent neural network,and recurrent neural network based on semantic analysis indicate that.The proposed method can effectively detect malicious code that is not easily detected by traditional detection techniques.3.Designed and implemented a simple Android malicious code detection system.The system mainly includes functions such as algorithm selection,feature extraction,feature selection,and result display,successfully applying the algorithm studied in this article to malicious code detection in Android applications.Tests have shown that the system has certain advantages in flexibility,automation,accuracy,and other aspects.Main contributions: By analyzing apk files,binary files are converted into grayscale images.By using CNN as the basic framework and resnet as the backbone network,self attention mechanism is introduced to improve the accuracy of the model.By reading APK files,processing text information,and using an improved semantic recognition model,a malicious code detection algorithm based on semantic analysis is proposed.A machine learning based malicious code detection system for Android applications has been designed and implemented through the above algorithms.