In recent years, cyberattacks on humongous networks have resulted in irreversible damages and financial losses to many organizations and infrastructures. To some extent, an intrusion detection system (IDS) provides a corrective measure by recognizing and detecting any hostile actions within the network by monitoring and accumulating information about system behavior. In the present scenario where data is enormous and complicated, feature selection within data has become an integral feature of the IDS pre-processing. The optimal performance in IDS is achieved by extricating the most informative features of traffic data in order to reduce computing costs and enhance classification rate. The objective of this review paper is to study and critically assess state-of-the-art feature selection methodologies used in IDS. For this, the research papers from 2004 to 2021 have been taken into consideration for study and critical evaluation. This study comprises research articles from journals, conferences, book chapters from a multitude of reputable international and national publishers including Elsevier, IEEE, Springer, Wiley, ACM etc. In this paper, in-depth comparative evaluation of feature selection methodologies has been presented. Till date, a wide range of feature selection algorithms ranging from statistical analysis to deep learning methods have been proposed by various researchers. The goal of this study is to review the most imperative contributions relating to feature selection approaches to IDS in terms of minimizing unrelated and duplicate characteristics. A general taxonomy of FS methods that comprises FS process, methods, comparison of FS approaches along with its pros and cons has also been discussed. The accuracy rate of univariate filter selection approaches such as pearson correlation, mutual information, anova F-test is investigated using the NSL-KDD dataset. To enhance the accuracy rate of the multi-class classification of network attacks, experiments are performed using hybrid feature selection techniques. The hybrid approach comprises anova F- test (filter) and recursive feature elimination (wrapper) methods to select the optimal feature set. Finally, the study analyzes challenges, possible remedies and future possibilities for feature selection methodologies to provide insights on current research trends in the area of IDS.