Research On Forecasting Of Malware Infection In Mobile Computers

The vigorous development of the Internet era has brought about a more complex network environment,and the situation of computer infection with malware has become increasingly complex and frequent,thereby hindering the development of the information industry,endangering citizen’s privacy and national security.The probability of mobile computers being infected with malware,as the most commonly used computer,is also increasing day by day.Therefore,how to accurately and quickly determine whether a computer is infected with malware is an important issue in current computer security research.At present,traditional methods for detecting malware often start from its specific behavior and are overwhelmed by the constantly changing malware.The rapid development of big data and machine learning technology provides a solution for accurately and quickly verifying whether computers are infected with malware.This thesis selects the open source dataset of the Kaggle platform,extracts effective information,and establishes a model to predict whether mobile computers are infected with malware.It analyzes and studies the important factors that affect the infection of mobile computers with malware.Firstly,a two-step filtering was performed on the dataset to extract relevant data for mobile computers from the original dataset.Then,mobile computer data with battery cycles in the range of [10,500] was selected based on the battery characteristics of mobile computers,achieving simplification from 16 million data to 746959 data.Secondly,data preprocessing,visual analysis and feature engineering are carried out on the filtered mobile computer data sets,including the processing of redundant features,features with missing values,outlier and similar values,data coding,feature expansion and feature screening.Feature expansion expands the number of features in the preprocessed dataset from 82 to 130.Feature filtering uses three methods: regularization filtering,Permutation Importance filtering and Boruta algorithm filtering to summarize and filter,and finally 93 important features are screened.Then,using the preprocessed and feature engineered dataset,three models,Light GBM model,FTRL Proximal model,and XDeep FM model,were established to predict whether mobile computers are infected with malware,and the models were evaluated and optimized.Comparing the evaluation indicators of the three models,it was found that the XDeepFM model had the highest AUC value,reaching 0.72.Therefore,it has the best predictive effect on mobile computer infection with malware.Finally,using the established models to obtain the importance ranking of feature variables,the Permutation Importance idea is applied to the FTRL Proximal model and XDeep FM model.After summarizing and analyzing the important factors that affect mobile computer infection with malware,corresponding countermeasures and suggestions are proposed.