An Easy-to-Classify Approach for the Bot-IoT Dataset
Joffrey L. Leevy, John Hancock, Taghi M. Khoshgoftaar, Jared M. Peterson
Abstract
Bot-IoT is a recent and publicly available dataset that represents botnet attack traffic in Internet of Things (IoT) networks. About 9,000 of the roughly 73,000,000 instances in the dataset are labeled as normal traffic. In this study, we provide an easy-to-learn approach for Bot-IoT. Our method involves the use of a minimum number of dataset features and a simple learning algorithm for accurate classification. To be more specific, our contribution revolves around the use of only 3 out of the 29 Bot-IoT features and the Decision Tree classifier. In keeping with our definition of easy-to-learn, we require that predictive models have Area Under the Receiver Operating Characteristic Curve (AUC) and Area Under the Precision-Recall Curve (AUPRC) mean scores greater than 0.99. Our results demonstrate that the Bot-IoT dataset yields an easy-to-learn Decision Tree model.