Research On Android Malware Detection Technology Based On Machine Learning

The fast advancement of mobile communication devices has aided the progress of social informatization significantly.At the same time,it has piqued the interest of numerous virus creators.Despite the fact that a growing number of researchers are conducting in-depth research on Android malware detection technology,which has effectively slowed the spread of malware,360 security brain has detected up to 1,809,000 new malware in 2019,indicating that the mobile security threat remains severe.In light of the aforementioned mobile security issues,this article conducts extensive research on current Android malware,outlines common malware behavior,and investigates Android malware detection on this basis.This thesis’ s key contributions and work are as follows:1)This thesis proposes an application complexity partition-based detection approach.The impact of application complexity characteristics on Android malware detection is discovered through research and experimental verification of existing detection technology,and a detection model based on application complexity division is proposed based on the application complexity characteristics.The original data set is divided into four data sets containing applications of various application complexity categories;in order to improve the efficiency of malware detection,a feature selection method based on frequency difference is proposed to select the most feature subset;and five commonly used machine learning algorithms are selected based on research on a variety of machine learning algorithms.The implementation of the application complexity partition model on the SVM classifier is effective,according to experimental verification.The best results are attained,with accuracy ranging from 95.18 % to 99.19 % and recall ranging from 95.45 % to 99.68 %,considerably enhancing malware detection efficiency.2)A malware detection model based on multi label classification is proposed in order to investigate the impact of application complexity on Android malware detection and the link between application complexity and classification outcomes.The impact of application complexity on malicious software detection is effectively solved,and the accuracy of Android malware detection is greatly improved,by converting the application complexity category into a label,the original binary classification problem into a multi label classification problem,and using the multi label classification algorithm to explore the binary association between the application complexity category label and the classification label.The accuracy of the calibrated label ranking algorithm,which is based on the second-order strategy multi label classification method,is 98.45%.This research examines the features of application complexity in depth using the studies mentioned above.This study proposes the corresponding detection model to avoid the interference of application complexity on Android malware detection accuracy,and substantially increase the accuracy and efficiency of Android malware detection.