A comparative study of machine learning algorithms for intrusion detection in IoT networks

Abstract

The pervasive threat of cyberattacks jeopardizes the security and privacy of the Internet of Things (IoT) landscape, spanning devices to networks. To counter these attacks, research has been directed towards the development of effective and appropriate countermeasures. Intrusion Detection Systems (IDSs), particularly those leveraging Machine Learning (ML) techniques for expedited attack detection, are currently recognized as some of the most potent solutions for preserving the integrity of the IoT environment. This study was conducted with the objective of evaluating the efficacy of supervised Machine Learning techniques, specifically, Random Forest (RF), Decision Trees (DT), and XGBoost classifiers, in detecting attacks within the IoT network. Chi-square (Chi2) and Mutual Information served as the employed Feature Selection Techniques. The research utilized two recent datasets for model evaluation. In pursuit of an optimal solution and high IDS model accuracy, a comparison of different techniques was undertaken across each stage of the ML workflow. The performance of the algorithms was assessed using the Edge-IIoT and BoTNeTIoT datasets, and the results from the two were compared. The impact of each workflow step on the model s accuracy was also examined. According to the performance metrics, the best results were achieved with the Mutual Information and XGBoost combination, outperforming both the Random Forest and Decision Tree classifiers. This study thus contributes to the ongoing efforts to strengthen IoT security through enhanced intrusion detection techniques.