A weighted voting-based ensemble classifier based on manta ray foraging optimizer for cyberattack detection in IoT environments: a comparative study

Recently, several machine learning (ML) and deep learning (DL)-based methods have been introduced for detecting cyberattacks in IoT networks. However, most of them are tested with only one or two datasets, which does not guarantee that the model will perform well on others with different characteristics or in various real-world scenarios. Therefore, this paper presents a comparative study of ML and DL models on five common network intrusion datasets: UNSW_NB15, NSL_KDD, Edge-IIoTset, RT-IoT2022, and IoTID20. Additionally, two new ML techniques are introduced: ET-MRFO and VC-RXE-MRFO. ET-MRFO uses the recently proposed manta ray foraging optimizer (MRFO) to tune the hyperparameters of the extra trees classifier, leading to more accurate detection of diverse cyberattack patterns and improved overall accuracy. VC-RXE-MRFO, a new weighted voting classifier, combines the strengths of three ensemble learning models—random forest (RF), extra trees (ET), and XGBoost. It also utilizes the MRFO algorithm to optimize the hyperparameters of RF and ET, as well as assign weights to various base models to emphasize stronger ones and mitigate the influence of weaker ones, resulting in a robust classifier that can detect cyberattacks more accurately. The models are validated and compared across five datasets using several performance metrics, including F1-score, precision, recall, and accuracy. Empirical results reveal that the proposed VC-RXE-MRFO surpasses all other models on most datasets, with an accuracy of 0.995780 on UNSW_NB15, 0.999936 on NSL_KDD, 0.993838 on Edge-IIoTset, 0.998834 on IoTID20, and 0.999526 on RT-IoT2022. Additionally, the findings indicate that ML models outperform DL models across all the datasets studied, highlighting their strong ability to detect cyberattacks in IoT environments.