Improving Machine Learning Models for Malware Detection Using Embedded Feature Selection Method

Abstract

Machine learning performance always rely on relevant phase of pre-processing, that includes dataset cleaning, cleansing and extraction. Feature selection (FS) is a crucial phase too, because it is intended to increase the efficiency of Machine Learning (ML) models in terms of predictiveness, by assigning a representative value to the most important features in a dataset of malware. In this study, we focus on feature selection using embedded-based methods in order to minimize computational time and complexity of ML models. Embedded-based methods combine advantages of both filter-based and wrapped-based methods, in terms of studying the importance of features while executing the model and their reduced time of execution. Applying ML models shows a high stability of models will selecting 10 most relevant features from the dataset, with an accuracy that achieve 99.47%, 99.02% for respectively Random Forest (RF) and XGBoost (XGB).