Research On Android Malicious Program Detection Based On Deep Learning

As the Android operating system occupies an increasingly important market position on the smart mobile terminal platform,malicious applications targeting the Android platform have become more and more common.In order to protect user privacy and data security,a malicious program detection method oriented to the Android system is needed,which can automatically detect unknown applications by combining multiple levels of information.Based on the above-mentioned concepts,this thesis proposes a method that combines dynamic and static characteristics of applications for Android malicious program detection.The method extracts the static feature vector and dynamic behavior log of the target application by analyzing the source code information of the application and monitoring the application system function call behavior.At the same time,a composite text processing neural network CNN-Bi LSTM-Attention model is used to abstract and convert the dynamic behavior log text into dynamic feature vectors.Finally,choose the deep learning network MLP model to complete the classification and detection of malicious programs.Combining dynamic and static features for Android malicious program detection utilizes information at multiple levels of the application,which helps improve the accuracy of detection.At the same time,the use of deep learning models to abstract text information into feature vectors,and to abstract and learn more information from the feature vectors,can detect more hidden malicious behaviors of the application.This thesis implements the proposed method into an Android malicious program detection system,and tests and experiments on the Android application data set collected by itself.The experiment is designed with different feature selections,different experimental parameters and different comparison models to verify the influence of these variables on the accuracy of system detection,and select the best system variables based on the results.Finally,the experimental results show that the average detection accuracy of the Android malicious program detection system implemented in this thesis is between 92% and 94%,which verifies the high usability and effectiveness of the method and system.