A Network Intrusion Detection Approach Using Extreme Gradient Boosting with Max-Depth Optimization and Feature Selection

Network intrusion detection system (NIDS) has become a vital tool to protect information and detect attacks in computer networks. The performance of NIDSs can be evaluated by the number of detected attacks and false alarm rates. Machine learning (ML) methods are commonly used for developing intrusion detection systems and combating the rapid evolution in the pattern of attacks. Although there are several methods proposed in the state-of-the-art, the development of the most effective method is still of research interest and needs to be developed. In this paper, we develop an optimized approach using an extreme gradient boosting (XGB) classifier with correlation-based feature selection for accurate intrusion detection systems. We adopt the XGB classifier in the proposed approach because it can bring down both variance and bias and has several advantages such as parallelization, regularization, sparsity awareness hardware optimization, and tree pruning. The XGB uses the max-depth parameter as a specified criterion to prune the trees and improve the performance significantly. The proposed approach selects the best value of the max-depth parameter through an exhaustive search optimization algorithm. We evaluate the approach on the UNSW-NB15 dataset that imitates the modern-day attacks of network traffic. The experimental results show the ability of the proposed approach to classifying the type of attacks and normal traffic with high accuracy results compared with the current state-of-the-art work on the same dataset with the same partitioning ratio of the test set.