Publication

Abstract : The increasing vulnerability of systems and the rise in malicious events have sparked concerns about network security. In order to address these threats, network intrusion detection systems (NIDSs) play a role in protecting against malicious threats. However, IDSs often face obstacles, like the issue of imbalanced classes, which can hinder the effectiveness of machine learning models by giving preference to the majority class. To resolve this issue, many strategies such as resampling, cost-sensitive, and ensemble learning systems have been proposed, but no relevant metrics have been developed to investigate the influence of observed performance on the data-level approach. The proposed model introduced a new metric to study the impact of sampling for the classification algorithm. This paper presents a novel approach known as the CIIR (Causal InferenceImbalanced Ratio) by utilizing ADASYN-IHT with Boruta-ROC feature selection in conjunction with four well-known imbalanced datasets: CIC-DDoS2019, UNSW-NB15, ML-EdgeIIoT and WUSTL-IIoT2021. The experimental outcomes prove the efficacy of the ADASYN-IHT and Boruta-ROC methods in improving classification performance on these datasets and by studying the impact of the CIIR.