A Novel Temporal-Aware Adaptive Feature Selection Strategy for Network Intrusion Detection Systems

n high-dimensional network intrusion detection, effective feature selection is crucial for robust performance and real-time detection. However, the conventional feature selection techniques lack exploration of temporal time-dependent features. In this paper we present a comparative study of a novel Temporal-Aware Adaptive Feature Selection (TAFS) method against conventional feature selection techniques like Mutual Information (MI), ANOVA, PCA,and LASSO in a binary classification scenario using the CIC-IDS 2017 dataset. TAFS integrates mutual information, feature importance score using Random Forests, and temporal analysis via Fast Fourier Transform (FFT) to capture patterns in time-dependent features. Experiments were conducted over four feature subsets (6, 8, 10, and 12) using a Random Forests (RF) classifier to evaluate the performance of the proposed TAFS method. Evaluation metrics include accuracy, precision, recall, F1-score (including weighted F1), receiver operating characteristic (ROC) curves, and confusion matrices. The results indicate that TAFS generally outperforms the conventional methods, demonstrating higher overall accuracy and improved ROC-AUC. These findings suggest that incorporating temporal characteristics into feature selection can significantly enhance intrusion detection performance.