High-dimensional datasets pose one of the main challenges that degrade the performance of IDS models, increase the classification models’ complexity and deteriorate the detection speed. Dimensionality reduction techniques aim to reduce the dimensionality of the dataset by transforming the original feature space to a representable low-dimensional feature space. Dimensionality reduction techniques are classified into feature selection and feature extraction. This study aims to find the smallest set of features that best represents the original feature space while achieving high classification performance. We also aim to find the feature selection technique that selects features that enable the fast detection of attacks and reduce the model complexity. The experiment evaluates the effect of using 4 feature selection techniques Chi-squared, ANOVA, Recursive Feature Elimination (RFE), and Extra-trees on Decision Tree (DT), Random Forest (RF), XGBoost (XGB), Naive Bayes (NB) in terms of performance metrics, training and testing times in iterative fashion. The experiment was conducted using the benchmark dataset CICIDS2017. This analysis shows that the combination of RFE and DT achieves the highest accuracy of 97.9% accuracy using 20 features, requiring the least training and testing time, reflecting the model’s simplicity and fast detection speed.