Effect of Features Extraction and Selection on the Evaluation of Machine Learning Models

The exponential growth of sophisticated malware attacks against computer systems, has alerted IT security experts on the shortcomings of traditional protection tools because, they became unable to detect new families of malware that are more advanced and use advanced tools such as polymorphism, metamorphism, and obfuscation tools. Nowadays, machine learning is widely used in several IT fields; and also in cybersecurity, and can be an essential tool for malware detection, moreover, it can go beyond the limits of classic malware detection methods, such as the signature-based method, the anomaly-based method and the hybrid-based method, etc. The purpose of this study is to analyze the feature selection and extraction effects on the performance of malware classification model using machine learning. The results show that reducing dimensionality of datasets can help to improve the efficiency of security models in restricted time but with high performance, Random Forest using chi-square achieves an accuracy of 99.51%. © 2022 Elsevier B.V.. All rights reserved.