Comparison of Machine Learning Algorithms for Malware Detection Using EDGE-IIoTSET Dataset in IoT

Abstract

The growth of IoT devices has presented great vul- nerabilities leading to many malware attacks. Existing IoT mal- ware detection methods face many challenges; including: device heterogene-ity, device resource restrictions, and the complexity of encrypted malware payloads, thus leading to less effective conventional cybersecurity techniques. This study's objective is to reduce these gaps by assessing the results obtained from testing five machine learning algorithms that are used to detect IoT malware by applying them on the EDGE-IIoTSET dataset. Key preprocessing steps include: cleaning data, extracting features, and encoding network traffic. Several algorithms used these include: Logistic Regression, Decision Tree, Na¨ıve Bayes, KNN, and Random Forest. The Decision Tree model achieved perfect accuracy at 100%, making it the best-performing model for this analysis. In contrast, Random Forest delivered a strong performance with an accuracy of 99.9%, while Logistic Re- gression performed at 27%, Na¨ıve Bayes at 57%, and KNN with moderate performance. Hence, the results have shown the effectiveness of machine learning techniques to enhance the security IoT systems regarding real-time malware detection with high accuracy. These findings are useful input for policymakers, cybersecurity practitioners, and IoT developers as they develop better mechanisms for handling dynamic IoT malware attack incidents.