A machine learning-based approach for multi-class intrusion detection and classification in IoT using CICIoT2023 dataset

The Internet of Things (IoT) is a set of physical devices connected to the Internet that collect information from the operating environment and share it with users or other devices. One of the most important challenges in IoT is the security of connected devices; they are vulnerable to attacks and infiltrations and traditional Intrusion Detection Systems (IDS) often fail to overcome this problem. In this article, six machine learning models, including Logistic Regression, AdaBoost, Perceptron, MLP, Random Forest and Hist-Gradient Boosting, are used to solve this problem. These models are trained on the recently available CICIoT2023 dataset and are used to classify attacks for three different classification modes of binary, eight and 34 classes. The performance of the models is evaluated based on several evaluation criteria such as accuracy, precision, recall, f1-score and training time. Evaluation results suggest that the Random Forest algorithm achieves the best classification results with test accuracy $=0.9955$, while the Hist-Gradient Boosting algorithm has the best performance in terms of training time.