Research on intrusion detection optimization methods based on feature selection
Yufei Wu, Zhe Wang, Kai Liu, Xiumei Wen · 2024
As the Internet increasingly penetrates into people's lives, network attacks should also receive relative attention, and intrusion detection is an important part of securing network systems. Traditional intrusion detection techniques have several drawbacks including poor accuracy, high false alarm rate, susceptibility to attacker spoofing and inability to handle complex attacks. In this study feature selection algorithms for intrusion detection are selected among others. XGBoost, DT and RF are used as base classifiers and XDR algorithm is designed to be trained on intrusion detection data using GB classifiers as meta-model. Stacked algorithms in the field of intrusion detection are optimized for improving the performance and accuracy of cybersecurity intrusion detection systems aiming at identifying various types of cyber-attacks more accurately. The algorithms in this paper are validated on KDPCAP and DARPA datasets. The algorithm performs better in terms of accuracy and F1 score with up to 2.82 percentage point increase in recall compared to the traditional Stacking algorithm on KDD and DARPA datasets.