OSE-IDS: Optimized Stacked Ensemble Intrusion Detection System Using Automated Machine Learning Approach

Abstract

As cybercrimes are increasingly evolving, the existence of an intelligent Network intrusion detection system (NIDS) is indispensable in the network infrastructure. In addition, there are many challenges face NIDS design based on Artificial Intelligence - technology such as irrelevant features in network traffic, rare examples of malicious traffic, and the efforts for Machine learning model selection and models' hypermeters finetuning. This study proposes efficient NIDS concerned with these challenges to accurately detect malicious behaviors. First, A parallel hybrid feature selection approach filters the most important features. Second, to address data imbalance, we integrated a combined Random Under-sampling Strategy and Synthetic Minority Oversampling Technique-Edited Nearest Neighbors technique to ensure the balanced representation of minority attacks. Finally, the stacked ensemble classifier, comprising the four best base models selected through the Automated Machine Learning approach. Using the CICIDS2017 dataset, a comprehensive benchmark for intrusion detection research, our approach achieves an impressive detection rate of 99.76%, effectively identifying both majority and minority classes.