Evaluating machine learning approaches for multiple attack classification with improved computational efficiency in IoT networks

The rapid expansion of Internet of Things (IoT) devices has introduced significant security challenges, making IoT networks increasingly vulnerable to cyber-attacks. This research study develops and evaluates a comprehensive approach for detecting and classifying Denial of Service, Distributed Denial of Service, and Mirai attacks in IoT environments using five supervised machine learning algorithms: Random Forest, Gradient Boosting, Naive Bayes, Decision Tree, and K-Nearest Neighbors, applied to the CICIoT2023 dataset. Our methodology incorporates data preprocessing techniques, including undersampling for class imbalance, and implements three feature selection methods: Chi-square, Principal Component Analysis, and Random Forest Regressor. We thoroughly evaluate model performance by measuring accuracy, precision, sensitivity, and F1-score metrics. In addition, we assess computational efficiency by considering training and prediction times. Our results demonstrate state-of-the-art performance, achieving the highest reported accuracy of 99.99% for the CICIoT2023 dataset, surpassing results reported in existing literature. Furthermore, our best-performing Decision Tree model exhibits significantly improved computational efficiency, with a 98.71% reduction in training time and a 99.53% reduction in prediction time compared to previous studies, while maintaining superior accuracy. This research contributes valuable insights into the effectiveness and efficiency of various machine learning approaches for IoT security, offering practical implications for developing robust and computationally efficient intrusion detection systems in resource-constrained IoT environments.