An Android Malware Detection Method Based On Deep Learning Of Dynamic And Static Features

Intelligent mobile terminals and the Internet are widely used today.Malwares are also moving to mobile terminals.Malwares on the Android platform spread so widely due to the extensibility and openness of Android system.In order to ensure the information security of mobile terminals,we need a reliable and effective method of Android malicious code detection.Of all the related detection technology,most methods are based on static analysis or dynamic analysis.Static analysis performs poorly in the malwares which are encrypted or loaded dynamically,while dynamic analysis is difficult to cover all malicious behavior.In order to solve the problems,this paper proposes an Android malware detection method based on dynamic and static features.Firstly,we get dynamic features by taint analysis,as well as we get static features by static analysis.Secondly,we use deep learning model to detect malicious code by the features we have got.The dynamic features and static features are combined,so that the behavior of Android applications can be better obtained,and the accuracy of malicious code detection is approved.By the usage of the deep learning model,the features can be abstracted to a higher level so as to detect the concealed malicious code.This paper makes an implementation of the method.The system is tested on the open Android application sample set.This paper designs experiments to analyze the effects of different features,different model parameters and different neural network models.Experiments show that the detection accuracy is between 85% and 95%.The method mentioned has a strong practicability,and it can be used to protect the information security of Android system.