A comprehensive evaluation of multiclass imbalance techniques with ensemble models in IoT environments

Abstract

The internet of things (IoT) has revolutionized connectivity and introduced significant security challenges. In this context, intrusion detection systems (IDS) play a crucial role in detecting attacks in IoT environments. Bot-IoT datasets often face class imbalance issues, with the attack class having significantly more samples than the normal class. Addressing this imbalance is essential to enhance IDS performance. The study evaluates various techniques, including imbalance ratio techniques we call imbalance ratio formula (IRF) for controlling imbalance data, while also testing IRF to compare it with oversampling techniques like synthetic minority oversampling technique (SMOTE) and adaptive synthetic sampling (ADASYN). This research also incorporates the extreme gradient boosting (XGBoost) ensemble model approach to improve IDS performance in dealing with multiclass imbalance issues in Bot-IoT datasets. Through in-depth analysis, we identify the strengths and weaknesses of each method. This study aims to guide researchers and practitioners working on IDS in high-risk IoT environments. The proposed IRF, when integrated with the XGBoost algorithm has been demonstrated to achieve comparable accuracy of 99.9993% while reducing the training time to be on average at least two times faster than those achieved by the other state-of-the-art ensemble methods.