Publication

Abstract : Botnets are one of the most prevailing threats for cyber-physical devices around the world. The evolution of botnet attacks has been rampant and diverse with vast scalability. One of the variants is targeting the IoT ecosystem involving devices not limiting to sensors, actuators, and all kinds of smart devices. Modern-day botnet threats have multiple functionalities rather than targeting devices for DDoS. In this paper, we used the two latest IoT Botnet data sets: IoT-23 and MedBIoT, which consists of modern-day attacks that helped us classify them for more than two classes. We have considered 6 variants of IoT botnet attacks from both the data sets and categorise them into 3 classes. We have used ensemble approaches for multi-class classification where random forest outperformed with an accuracy of 99.88. We have also generated new samples using conditional generative adversarial networks (CTGAN) for testing the efficacy and robustness of our models built.