To solve the problems of too many features and insufficient accuracy in Android malware detection,a feature impor-tance scoring algorithm was proposed.The lack of weight reduction of common features in term frequency and inverse document frequency was improved.Three types of static features of permission,intention and interface were extracted.The diverse fre-quency and distribution differences in benign and malicious data sets were extracted,and they were ranked successively according to the resulting score size.The key features with more differentiation were screened out.Experimental results show that,combi-ning the first 150 key features selected using this method with random forest model,the accuracy reaches 98.82％that is superior to other algorithms under the same conditions,which meets the needs of practical application.
维普
万方数据