A feature exploration approach for IoT attack type classification
Masoud Erfani, Farzaneh Shoeleh, Sajjad Dadkhah, Barjinder Kaur, Pulei Xiong, Shahrear Iqbal, Suprio Ray, Ali A. Ghorbani
Abstract
Internet of Things (IoT) devices' unique identity and adequate network infrastructure of physical objects em-bedded with software, actuators, and sensors create an open playground for various cybersecurity attacks. Recently, several researchers attempted to simulate diverse datasets to mimic the behavior of IoT devices and potential attacks in this field. However, since more new and dangerous attacks are being produced, a more diverse and universal dataset is required in this field. This paper proposes a framework to enrich the IoT datasets in two directions: Vertical and Horizontal. The Vertical aspect merges famous state-of-the-art IoT datasets, and in the Horizontal aspect, we propose a unique and new set of features to present the behavior of IoT devices in more diverse settings. Our experimental results demonstrate that the new simulated datasets enhanced by our method have achieved better performance in classifying cybersecurity attacks with various machine learning algorithms. All the generated datasets and codes created for this paper are publicly available in https://www.unb.ca/cic/datasets/enricheddataset.html.