EnFeSTDroid: Ensembled feature selection techniques based Android malware detection

Abstract

Android smartphones have gained widespread popularity since 2008, making them frequent targets for malware. To address these threats, researchers have developed various detection models. Most existing techniques either use a single feature selection technique or combine a very few selection techniques, which can lead to overlooking other important features. In this study, we propose a novel method that first extracts permissions from applications. Then, it applies six different feature selection techniques, namely Information Gain, Extra Tree Classifier, Chi-Square, Mean Term Frequency (MTF), Inverse Document Frequency (IDF), and Mean Term Frequency–Inverse Document Frequency (MTF–IDF), to rank the permissions from the most to least significant. Furthermore, it applies Friedman’s and Post hoc Nemenyi tests to combine the rankings and identify the most relevant and distinguishing features for classifying malware. The results show that our proposed model could accurately classify 96.27% of the samples. Our work is novel and significant, as we have combined six feature selection techniques to enable the model to leverage the advantages of all the methods, rather than relying on a single or a few techniques. The proposed work also outperforms several other existing works in the literature.