Feature Selection in Cybersecurity: A Comparative Study of Machine Learning Models

Abstract

Machine Learning (ML) models have proven their utility and benefits in the field of cybersecurity, a domain that requires confidence and reliability as essential criteria for success. While ML algorithms are capable of meeting these requirements, they are not without limitations. Issues such as high temporal complexity due to the vast number of features, performance bottlenecks, and difficulties in interpreting the results persist. Large-scale data analysis remains a significant challenge in ML, particularly in Cybersecurity. Large-scale data analysis is always a challenge in the ML field. Feature Selection (FS) provides an effective way to solve this problem by eliminating irrelevant and redundant data. This process not only reduces computation time but also enhances learning accuracy and facilitates a better understanding of the learning model and data. This paper highlights various existing FS methods, detailing the metrics employed and the types of these methods, such as filter models, wrapper models, and embedded models, applicable to both supervised and unsupervised learning approaches. First, we will present a literature survey on FS methods applied in cybersecurity, including approaches based on information theory, probability, genetic algorithms, and meta-heuristic techniques. Then, we discuss and categorize these methods within the context of cybersecurity, leading to a synthesis on their impact in enhancing both the performance and interpretability of the models used in this field.