Cybersecurity uses Stacking Ensemble Learning for Darknet Classification

Abstract

Cyber intelligence services sometimes refer to the "darknet," the part of the internet that consumers don't often anticipate to be accessible for machine-to-machine communication. Before building security, it is important to analyze the network's risks. To examine and classify darknet data, this study suggests new classification methods for machine learning called stacking ensemble learning. This study used ensembles of machine learning methods on the newly published CIC-Darknet2020 dataset to accurately distinguish between Darknet and Benign traffic, achieving a 98% accuracy rate. Furthermore, it successfully identified the specific kind of application running behind the Darknet traffic with a 97% accuracy rate. In addition, we used an approach based on game theory to assess the output of the models developed using machine learning and showcase the impact of the features, intending to gain a deeper understanding of the Darknet traffic behavior. To the best of our knowledge, this research is the first one conducted on this dataset, as confirmed by the dataset producers.