Study And Implementaion Of Malware Behavior Detection Of Android Software Based On Deep Learning

With the rapid development of mobile Internet, smartphone has gradually become an indispensable device for users to use the network and plays an almost indispensable role in people's daily life. Android system become the world's highest market share of smart phone operating system and attracts major manufacturers, developers, and users of all ages with its opensource feature. Due to the opensource feature of Android system and bad code review mechanism for downloading applications from third-party application market, it made the Android market having a lot of malicious applications. Android users download and install Android applications with malicious behavior, it is a threat to the mobile phone system and the user's privacy data threat (such as system failure, theft of privacy, malicious deductions). Therefore, how to detect a malicious Android application, it is very important for the Internet security,application market, or the Android users.Based on the analysis of current research situation at home and abroad, this paper deeply studies the malicious software behavior process of Android software and proposes a method for Android software malicious behavior detection based on deep learning. The main research focuses include:First of all, analyze the current Android malicious behavior categories, the implementation process and key detection techniques,summarize Android software detection technology, the general steps of the process and the required technology.Then, analyzed the text modeling method and the advantages and disadvantages of word vector technology. And then proposed a method of extracting malicious feature of Android software based on word vector.Last but not least, analyzed the advantages and disadvantages of the classification algorithm recursive neural network based on LSTM. And then extract the feature using the word vector. Then, establish the malicious behavior detection model of Android software by the recursive Android software, Experimental results show that the proposed algorneural network based on LSTM. Finally used this model to detect unknown ithm can achieve better results than traditional machine learning algorithms.