Research On Android Malware Detection Technology Based On Deep Learning

As Android smart phone is growing closely related to everyone's life,its security problems have become increasingly prominent.There are many efforts to address the security problem on Android platform.However,most approaches are based on traditional machine learning which can not understand the semantic feature information of more comprehensive Android malware.In response to many hidden dangers brought by severe security problems,especially the Android malware,a new Android malware system based on deep learning is proposed in order to detect Android malware more effectively.The main research steps of the system are as follows:firstly,the benign applications collected by web crawler technology and the malware applications are collected from the research institutions to form data sets for experiment.Secondly,the static detection technology and dynamic detection technology are used to extract the relevant Android security features.Thirdly,the extracted feature set is preprocessed.Fourthly,in order to recognize different Android malware,different kinds of clustering algorithms can be applied to compare the malware modeling capability.Fifthly,the best classifier is selected by using various evaluation indexes.The comparison of the experimental results led us to find that the deep learning is more accurate in the Android malware detection than the traditional method.Finally,a more efficient combination classifier built on experimental training is used to implement a system,which is called DroidGuard.The main work of this thesis is that compared with the traditional approach,our study adopts deep learning to study the classification of Android malware since deep learning can not only improve the security semantic information of the Android software,but also increase the capability of the classifier model to learn and comprehend the security feature of the software.Furthermore,a better combination classifier is constructed for 49662 applications in reality.The accuracy of the system is found to be up to 97.9%,a higher degree of accuracy and credibility compared to related research.Experimental results show that the proposed method can effectively detect Android malicious software and has practical application value.