Feature engineering plays most significant role for an efficient intrusion detection system. To optimize the neural networks performance, includes the careful selection and transformation of input features. This review article provides a comprehensive study of FE in the IDS, and signifies the importance of feature selection and dimensionality reduction. The importance of FE in enhancing model performance, mitigating challenges related to high dimensionality and complexity of network data, and addressing the dynamic nature of cyber threats is highlighted. Moreover, feature selection has various methods like filter, wrapper, embedded methods used for feature selection and dimensional reduction methods like PCA, an unsupervised technique, mostly used for DR, and suitable for all types of datasets, whereas t-SNE, the most newly evolved non-linear approach get better result on large datasets with large number of features. LDA is a linear supervised technique, that performs better on huge datasets by classifying multiclass datasets. This article outlines the significance of effective feature engineering to speedup model training, optimized use of computational resources, and enhanced model overview in IDS for ML and DL.