Comparing Machine Learning and Deep Learning for IoT Botnet Detection

Botnet has become a major threat to Internet of Things (IoT) devices due to the low security settings from manufacturers and the lack of security awareness from end users. Many ports are open by default and default user credentials are left unchanged. To solve the increasingly popular botnet attack, many detection approaches have been proposed. However, most of them are targeting on one particular approach or one botnet dataset. There is lacking a comprehensive comparison between different machine learning and deep learning approaches on this task under different datasets collected from different ways. One of the main areas of study about the botnet attack is the comparison about different datasets and how different machine learning algorithms and deep learning algorithms are able to detect the difference between traffic data. In this work, we have measured the performance of 5 machine learning and 2 deep learning-based approaches on 4 recently published IoT botnet datasets collected using real and virtual IoT devices under Mirai malware attack. Our comparison results have shown that decision tree achieved the best detection accuracy as well as the shortest training and testing time.