An Unsupervised Machine Learning Approach for IoT Device Categorization

Internet of Things (IoT) along with the advances in the recently emerged Edge Computing environment, have allowed the introduction of new and very diverse applications that can facilitate our everyday life. However, one intrinsic characteristic of IoT is the heterogeneity of the IoT devices that are continuously connected and disconnected, creating a highly volatile communication environment. In addition to that, new types of IoT devices are constantly manufactured making a supervised categorization approach not applicable due to the lack of historical data. Nonetheless, the classification or type identification of the IoT devices is important for the management and the decision making of the IoT applications, and can be used for traffic characterization, density prediction, network planning and security reasons among others. Accordingly, in this paper we propose for the first time an unsupervised machine learning methodology for the IoT device categorization that leverages traffic characteristics obtained at the network level. To this end, we tackle the limitation of requiring an annotated dataset, while our model could also work efficiently with new and not previously detected IoT devices. To do so, we experimentally evaluate our approach using two clustering algorithms namely, the K-Means and the BIRCH in a real dataset. The experimental evaluation presents promising results that enhance the applicability of unsupervised approaches for the IoT device categorization problem.