Feature Selection Algorithm Characterization for NIDS using Machine and Deep learning

Data dimensionality is increasing at a rapid rate, posing difficulties for traditional mining and learning algorithms. Commercial NIDS models make use of statistical measures to analyze feature sets including packet length, inter-arrival time, and flow size, in addition to other internet traffic parameters. Emerging algorithms must deal with diverse data. While multiple deep learning-based solutions exist in the literature, their commercialization is still in its infancy. Currently available machine learning techniques create a large number of false positives. As the quantity of data to be processed has increased in recent years, feature selection (FS) appears to have become a basic requirement for any type of model. The recent advent of promising techniques and different kinds of features advances existing computational research and continuously improves feature selection, seeking to make it applicable to a broader range of applications. This paper intends to provide a fundamental investigation to feature selection throughout NIDS, which will take into account basic concepts, categorization of existing systems, a framework and taxonomy for NIDS, the feature selection methods used by researchers to develop NIDS methods, and a comparison of FS Algorithm classification and Python FS library contents. By examining existing contributions, this study provides an overview of the majority of techniques proposed in the feature selection research. Additionally, we discuss the most recent FS algorithms for NIDS that were established to select the optimal feature subsets. By classification and comparative study, the paper provides a road map for comprehending and constructing the current state of NIDS FS. As a result, a study is presented to help the reader comprehend the research progress, the FS Algorithm's characterization, and the establishment of a new taxonomy for emerging developments and existing challenges.