Evaluasi Performa Algoritma XGBoost dengan Teknik Feature selection Chi-Square dan Feature Importance pada Sistem Intrusion Detection Berbasis Dataset UNSW-NB15

Abstract

 The increasing complexity and volume of network traffic make Intrusion Detection Systems (IDS) essential for accurate and efficient cyberattack detection. This study evaluates XGBoost combined with feature selection techniques using the UNSW-NB15 dataset. Feature selection was performed sequentially using Chi-Square as a filter method, followed by XGBoost Feature Importance as an embedded method. A total of 14,280 data instances after cleaning were divided using Stratified Hold-Out Validation with ratios of 60:40, 70:30, 80:20, and 90:10. Three models were compared: XGBoost without feature selection using 47 features (Model 1), XGBoost with Chi-Square using 20 features (Model 2), and XGBoost with Chi-Square and Feature Importance using 10 features (Model 3). Evaluation using Accuracy, Macro Precision, Macro Recall, Macro F1-Score, and Confusion Matrix showed that Model 3 achieved its best performance at the 80:20 ratio, with 82.95% Accuracy, 0.6548 Macro Precision, 0.5975 Macro Recall, and 0.6049 Macro F1-Score. Training time was also reduced from 30.06 to 11.57 seconds. However, minority-class detection remained limited due to class imbalance.