A Consensus-Based Feature Selection and Classifier Benchmarking for Network Anomaly Detection

Hakimi, Rifqy
Shalannanda, Wervyan
Heriansyah

Journal of Engineering and Scientific Research (JESR)
Vol. 7 No. 1 (2025)

Efficient anomaly detection in network traffic is essential for securing modern digital infrastructures. This study presents a comprehensive comparative analysis of six feature selection methods—including Mutual Information, Recursive Feature Elimination (RFE), LASSO, Random Forest Importance, ANOVA, and Chi-square—and seven machine learning classifiers on the NF-UQ-NIDS-v2 dataset. Experimental results demonstrate that advanced feature selection methods, particularly Mutual Information and RFE, combined with ensemble classifiers such as Random Forest and XGBoost, achieve superior detection performance. A consensus analysis reveals that features like protocol type, packet length, and flow duration are consistently most informative for anomaly detection. These findings provide practical guidance for designing accurate and efficient intrusion detection systems in high-dimensional network environments.