A Robust Ensemble Learning Approach for Real-Time Intrusion Detection in Network Traffic

The Internet has permeated all aspects of our lives, offering various benefits, but it can be dangerous if misused. Network attacks pose an alarming challenge in today's Internetdriven environment. Intrusion Detection System (IDS) is a beneficial way to address this issue, as it detects malicious access within network traffic. The demand for Intrusion Detection Systems is growing with the increasing risk of cyber attacks in the network. Currently, many machine learning methods are being applied to Intrusion Detection Systems because they ensure predictive power by leveraging patterns from past data. However, most of the studies have adopted outdated datasets that lack attack diversity. Moreover, poor detection rates have been observed due to the use of weak models and the presence of irrelevant features. This research proposes an efficient Intrusion Detection System using the Information Gain feature selection method with a Hybrid ensemble model. The CSE-CIC-IDS2018 dataset has been utilized to train the model, consisting of real-time network data and several types of attacks. At first, the Information Gain process selected 51 significant features. Then the RF, ET, and XGBoost models have been integrated into the Hybrid ensemble model and tested on the selected features. The use of multiple ensemble strategies has boosted the performance of the proposed model, leading to 99.9994% accuracy. Furthermore, the Hybrid ensemble approach has outperformed all other recent research methods in a comparative analysis using various datasets (CSE-CIC-IDS2018, NSL-KDD, and UNSW-NB15). Based on the findings, the proposed Hybrid ensemble model can be considered an advantageous strategy for detecting attacks in the network.