A Data Sampling and Two-Stage Convolution Neural Network for IoT Devices Identification

The rapid development of Internet of Things (IoT) enables emerging user services and applications to improve life quality. However, the presence of rogue IoT devices can result in the vulnerabilities that hurt users. In order to address this threat, organizations often apply security policies in which only the connection of white-listed IoT devices is permitted. To obtain that goal, organizations must be able to identify the IoT devices connected to their networks and, more specifically, to identify connected IoT devices that are not in the white-list (unknown devices). However, for new/unknown devices, it is often difficult to collect enough data samples to train an effective detection model. To address this problem, we propose a model that combines a data sampling technique with a two-stage Convolutional Neural Network to identify IoT devices. The proposed model is called DS-2CNN. DS-2CNN can accurately identify the IoT devices with very little training samples, thus allow the model to early identify unknown IoT devices. We have carried out the extensive experiments on the imbalanced dataset, i.e., the IoT Trace Dataset including 22 IoT devices. The experimental results have shown that DS-2CNN can significantly enhance the accuracy in identifying IoT devices comparing to a recent proposed model.