Deep Iot: A Deep Learning Model For Anomaly And Botnet Detection In Iot Networks

Abstract

The Internet of Things (IoT) is currently transforming the world by connecting physical objects to the Internet. However, as the number of connected devices and the growth of IoT continue to rise, new network security threats are emerging due to vulnerabilities in these devices. One prevalent threat is the presence of bot malwares, which exploit vulnerable IoT devices to launch cyber attacks. To address these risks, there is a need for novel methods to detect IoT botnet networks. In this study, we propose a network intrusion detection model that utilizes deep learning, specifically an Autoencoder, to identify malicious botnet traffic. Our model takes a one-class classification approach, focusing on modeling the legitimate behavior of devices within the network to detect anomalies without requiring manual labeling. To analyze device behavior, our solution generates network flows from traffic data and selects relevant flow statistics. We evaluated our approach using the IOT-23 dataset, which includes captures of botnets executed on IoT devices as well as legitimate IoT device traffic. The results demonstrate that our detection model achieves a high predictive performance in identifying different types of botnets, with an impressive F1-score of 93%