Intelligent Intrusion Detection Using ML for Large-Scale IoT Networks

Today's IoT infrastructure is continuously expanding with the inclusion of various devices, protocols, components, and so on, providing cyber attackers with multiple entry points to breach. That is why network security is vital in the large-scale IoT environment. However, defending the IoT system from cyber threats is challenging. In recent years, network intrusion detection using ML algorithms has gained prominence. Therefore, the research employed and examined the performance of four ML classifiers: logistic regression (LR), perceptron, adaptive boosting (AdaBoost), and random forest (RF) using the CICIoT23 dataset. The CICIoT23 is a recent real time benchmark dataset for large-scale IoT. The research performed three types of classifications: Binary classification (2-class), Grouped classification (8-class), and Multi-class classification (34-class). The research reveals that RF outperforms the other three ML classifiers. The proposed RF classifier successfully identifies and classifies various attacks as follows: for 2-class, 99.75% accuracy and 94.89% F1-score; for 8-class, 99.64% accuracy and 99.63% F1-score; for 33-class: 99.52% accuracy, and 99.50% F1-score. Moreover, the performance of the proposed RF classifier is superior to existing state-of-the-art ML techniques.