Research On Android Malware Detection Based On Recurrent Neural Network

With the widespread use of smart phone devices,there are more and more personal core privacy information stored in mobile phones,and the benefits brought by devices and information are getting higher and higher,and attacks on mobile devices are becoming more and more obvious.Faced with the increasingly new Android malware,which reduces the threats caused by malware for mobile phone users and better protects the interests of users,the research on malware detection for Android has been highly valued.As machine learning technology continues to mature,it performs better in dealing with classification problems.However,the increasing types of malware have greatly increased the difficulty of machine learning detection.Compared with traditional machine learning,deep learning uses surface features to learn abstract deep features.The cyclic neural network connects the units between the hidden layers,and has a more significant ability to process the timing and semantic information of the data.Deep cyclic neural network is to increase the number of network layers on the basis of cyclic neural network to make the network have better expressive ability.The long and short-term memory network structure overcomes the problem of the lack of contextual information that the standard recurrent neural network has to deal with for a long time,and is explicitly used to design the problem of long-term dependence.Aiming at the problem of low accuracy of existing machine learning-based detection methods,this paper analyzes the characteristics of benign and malicious software and preprocesses the features of the Omni Droid dataset to obtain an efficient and accurate Android malware detection method.The main research contents of this paper are as follows:(1)This thesis uses Android software as feature information.After analyzing benign and malicious software,most of the features cannot effectively distinguish between benign and malicious software.Feature extraction of all features in the data set will cause a large amount of redundant data,increase the time of classification model,and reduce model detection.The accuracy rate.Therefore,feature selection is performed on the data set used in this article,and the accuracy of detection can be improved by reducing the feature dimension.(2)This thesis proposes an Android malware detection model based on a deep recurrent neural network.The five types of features are selected using the information gain algorithm respectively,and the features with larger information gain values are stored in a feature set to construct a deep recurrent neural network classification model.Experiments show that compared with other machine learning classification algorithms,the detection method based on deep recurrent neural network shows better accuracy.(3)This thesis proposes an Android malware detection model based on Bi-LSTM.The five features in the data set are selected by genetic algorithm,the selected features are combined to form a new feature set,and a bidirectional LSTM classification model is constructed.The experimental results show that the Android malware detection effect based on Bi-LSTM is higher than that of the deep recurrent neural network model.