IoT Network Cyberattack Detection Performance Using Ensemble Machine Learning Based Intrusion Detection System
Abstract
Heterogeneous architecture, large-scale data generation and susceptibility to variety of cyber attacks are some of the major factors that contribute towards security challenge with rapid expansion of Internet of Things (IoT) networks. Classic intrusion detection systems (IDS) face challenges to maintain high detection accuracy and robustness in such variable environments. In addition, this paper presents a novel approach towards an ensemble of heterogeneous machine learning and deep learning classifiers as an intrusion detection framework for IoT networks to improve the performance of valid attacks detection. The suggested model combined Decision Tree (DT), Random Forest (RF), XGBoost, and Convolutional Neural Network (CNN) with a confidence-based soft voting mechanism enabled by Mutual Information feature selection to reduce redundancy and computational complexity. In order to generalize across all attack types, a diversified combined dataset from CICIDS2017, NSL-KDD, IoTID20, CIC-IoT2023 and ransomware traffic is created that accurately represents modern IoT cyber threats. Performance of the system was assessed for both binary and multi-class classification modes based on conventional performance measures (accuracy, precision, recall, F1-score and ROC-AUC). The experimental results prove that the proposed ensemble model always outperforms individual classifier: It achieves an accuracy of 99.08%, precision, recall and F1-score are also similar to 99.07 in case of multi-class classification, and for the binary classification task it outperforms traditional machine learning models. The ensemble also shows great discriminative capacity with an ROC-AUC of 0.955, which confirms that the ensemble is robust in differentiating between normal and malicious traffic. The results validate the effect of integrating heterogeneous learners with effective feature selection in improving detection accuracy, stability, generalization capabilities in complex IoT environments. This practical foundation for safe AIoT systems lays the groundwork to expand upon, as we demonstrate a scalable and reliable IDS solution that can defeat various cyber-attack patterns.