AutoBots: A Botnet Intrusion Detection Scheme Using Deep Autoencoders

Recently, with the massive exchange of data over Internet of Things (IoT) ecosystems, attacks surfaces have also intensified. In IoT, connected devices share data over open channels and thus are highly vulnerable to security and privacy attacks. Botnet-based attacks have been found to have a significant effect on the network-based system. Thus, in this paper, we present a scheme AutoBots, which differentiates the normal and anomaly behaviour of IoT devices among the connected network. To exploit this, we consider diverse parameters like network behaviour profiles and apply autoencoders to classify and detect anomalous traffic from normal traffic. We used the BASHLITE and MIRAI IoT botnet setup and trained our network with the N-BaIoT dataset that has both benign and malicious network traffic. We compared our scheme for metrics like attack detection time, attack detection with respect to hourly traffic, deep residual accuracy, and residual loss. The presented results signify the efficacy of the proposed scheme against conventional bot-detection schemes.