Comparative Analysis of Machine Learning Algorithms for Intrusion Detection in IoT Networks

Abstract

The growing threat of cyberattacks poses significant risks to the security and privacy of the Internet of Things (IoT), affecting everything from devices to networks. In response to these threats, research has focused on developing effective countermeasures. Intrusion Detection Systems (IDSs), particularly those utilizing Machine Learning (ML) techniques for faster attack detection, are now recognized as some of the most powerful solutions for safeguarding the IoT environment. This study evaluates the effectiveness of various supervised Machine Learning techniques, specifically K-Nearest Neighbors (KNN), Random Forest (RF), Decision Trees (DT), and Support Vector Machines (SVM), in detecting anomalies within IoT networks. The performance of these algorithms was assessed using the BoTNeTIoT-L01-v2 dataset. The results show notable performance differences before and after normalization. SVM emerged as the top performer, achieving 100% validation accuracy both before and after normalization. In contrast, RF and DT classifiers saw improved accuracy after normalization, reaching 92%, while KNN’s accuracy decreased post-normalization. Additionally, clustering techniques, including K-Means and soft clustering, were used to categorize network behavior. By comparing multiple ML approaches, this study makes a valuable contribution to advancing IoT security through enhanced intrusion detection methods.