Analyzing Mobile Applications Behaviors Based On Machine Learning

Now the reason why mobile application behavior to identify more challenging,because most mobile applications in order to protect user privacy and security for mobile applications is encrypted.Now the mobile application specific behavior recognition and analysis technology based on encrypted traffic has become a hot topic in the field of machine learning.From the perspective of encrypted traffic,this paper is based on machine learning related knowledge and aims at the technologies used in mobile application behavior recognition analysis: data collection and cleaning,data preprocessing,and encrypted traffic algorithm modeling.Based on the original research,Improvements and innovations have been carried out,and the specific program content is as follows:(1)In response to the lack of specific behavioral data sets for mobile applications today,there are very few papers on mobile application behavior recognition.A method for automatically operating mobile phone applications is proposed,which realizes the collection of large-scale mobile application specific behavior data sets.The program automates the operation of mobile phones by writing automated test scripts for mobile applications,then use the packet capture tool to capture the data packet on the mobile phone,and realize the packet segmentation and marking data flow set.(2)Aiming at the problem that different behaviors in the application will produce similar data streams and affect the accuracy of the classifier,an interference sample filtering method based on spectral clustering algorithm is proposed.Finally,the random forest algorithm is used for modeling and analysis.Experiments show that this method reduces the problem of similar interference and improves the accuracy of classification and generalization ability.(3)In view of the difficulty in obtaining useful information from encrypted data streams,a feature extraction scheme based on Burst is proposed.Finally,the random forest algorithm is selected for modeling comparison.Experiments show that this method improves the classification accuracy.(4)Aiming at the problem of selection of algorithm modeling corresponding to different encryption protocols at this stage,XGBoost algorithm is proposed for modeling.Experiments show that this algorithm is very outstanding in accuracy and avoiding overfitting.The method proposed in this paper is used to complete the recognition and analysis of Android mobile application behaviors.Based on this,the classification algorithm has higher accuracy and generalization ability.Finally,the shortcomings of this method and the future research expansion directions are put forward.