Intersection Filtering based on Recursive Feature Elimination Cross-Validation to Improve Classification Models in Early Detection of Android Malware

Abstract

Android malware is malicious software designed to damage or steal data on Android operating system devices. Machine learning models can be a solution for the early detection of Android malware. The problem in machine learning is that the large dimensions of the malware dataset can cause the model performance to be less than optimal. In this research, the proposed method of Intersection Filtering based on Recursive Feature Elimination cross-validation (IF-RFECV) is used for the process, which is expected to create a model that is robust to several types of high dimensional data, especially Android malware detection datasets. The research results show that Intersection Filtering based on RFECV (IF-RFECV) can produce fewer features and correlate with the label or target class. Overall, feature reduction using Intersection Filtering based on Recursive Feature Elimination Cross Validation (IF-RFECV) can produce accuracy, precision, recall and f-1 scores on the classification model that are better than original features or RFECV alone. The processing time carried out by Intersection Filtering based on Recursive Feature Elimination cross-validation (IF-RFECV) is similar to original features or RFECV alone. With the increase in results, this model can be used well in detecting malware on the Android operating system.