Optimizing Network Intrusion Detection Systems for IoT: Insights from Ensemble and Neural Network Models

The rapid increase in IoT devices has resulted in a significant rise in network traffic, creating new challenges in maintaining strong cybersecurity. To effectively combat both known and unknown cyber threats, such as DoS attacks, advanced network intrusion detection systems (NIDS) in IoT environments must ensure real-time and thorough detection. This study investigates the effectiveness of various machine learning and deep learning models in identifying DoS attacks using the TON_IoT dataset, specifically focusing on its network traffic component. The employed models included ensemble classifiers like the Extra Trees Classifier and Gradient Boosting Classifier, as well as various neural network architectures, including the Multi-Layer Perceptron (MLP) Classifier, Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and Artificial Neural Networks (ANN). The experimental findings reveal that the Gradient Boosting Classifier achieved the highest performance with a flawless accuracy of 100%, while both the Extra Trees Classifier and MLP Classifier obtained commendable outcomes with accuracies of 90%. The CNN also performed well, achieving an accuracy of 95% due to its high precision and recall metrics. Conversely, both the DNN and ANN struggled with the complexity of the dataset, resulting in lower accuracies of 68% and 38%, respectively. In summary, the findings indicate that ensemble learning techniques, particularly the Gradient Boosting Classifier, are effective in detecting DoS attacks and contribute to the development of robust NIDS tailored for IoT contexts. Thus, this research adds valuable perspectives on creating efficient intrusion detection systems to address the distinctive security challenges commonly faced by IoT networks.