The Research Of Android Malware Detection Method On Deep Belief Networks

In recent years,because Android operating system has many unique advantages,including ease of use,open-source code and open-source code and high scalability,the smart device equipped with Android operating system has occupied the majority of market share of the mobile smart devices market.At the same time,the characteristics of Android operating system and potentially huge benefits makes the smart mobile devices become the target of many malicious applications.Therefore,in order to protect the integrity,availability and confidentiality of Android user information,a valid Android malware detection system is very necessary and urgent for mobile manufacturers and security vendors.Currently,the analytical method of Android application can be divided into: static analysis,dynamic analysis and machine learning analysis.Static analysis refers that decompiling Android application package file,and then parsing the source code,the advantage of static analysis is lightweight and fast,however,there may be code confusion and it can't simulate the behavior of the applications,it can lead to high false positive.Dynamic analysis can monitor the flow of sensitive data in real time,including high precision,but that often need to modify the Android system source code,and this heavyweight method requires larger system resource.In machine learning analysis,by pre-trained Android application classifier that can simulate the behavior of Android applications,directly inputting the feature that can be extracted from tested application into the classifier,then completing identification and detection,but for the shortcoming of smart mobile device included weak computing power,small storage space and poor endurance,facing these problems that selecting what kind of machine learning method and how to build the vector space of Android application' features,so far,there is not a perfect answer.Because multi nodes in hidden layer,the deep learning method has superior ability to learn the original features and get better classification results.Compared to the traditional machine learning method,deep learning method has made breakthrough in many areas,such as semantics recognition,but the research of Android malware detection method on deep learning is rare.Therefore,in this paper,based on the research progress that machine learning analysis in the area of Android malware identification and detection,firstly,analyzing the static features and dynamic features of Android application,then building vector space of the application's original features;secondly,analyzing the related theory of DBN(deep belief network),proposing DAODB model that the Android malware identification model based on DBN,and then designing AMDOD system that the Android malware layered detection system based on DAODB model;finally,by experiments,find that Android application classifier based on DAODB model reached 95.45% in accuracy,by comparison with the classifier based on traditional machine learning,it is more better,and also find that the classifier can achieve better classification results under the conditions of mixing dynamic and static features than under the conditions of single features.