Ensemble Learning Approaches for Enhanced Internet of Things Cyberattacks Detection: a Comparative Study

For the last few decades, the Internet of Things (IoT) has witnessed remarkable growth, improving the quality of human life in various domains. However, owing to their Internet accessibility, IoT networks are highly vulnerable to cyberattacks, which pose severe risks. Ensemble Learning techniques are commonly used for attack detection to ensure security. Employing a single algorithm is limited; therefore, integrating multiple machine learning algorithms into ensemble models offers a complex but accurate solution. This study evaluates ensemble techniques using diverse IoT datasets (N-BaIoT, BoT-IoT, IoT-23, and WUSTL-IIoT-2021) for multiclass classification. Different ensemble learning methods (voting, stacking, bagging, boosting, and blending) were combined with individual classifiers (Random Forest, Decision Tree, k-nearest Neighbor, and Logistic Regression). The results demonstrate the importance of selecting an ensemble method that aligns with specific dataset characteristics. This study presents helpful insights for optimizing IoT project accuracy by understanding accuracy variations across datasets and ensemble models.