Traffic Behavior-based Device Type Classification

Abstract

Network operators play a fundamental role of providing network connectivity which fulfill service requirements of user devices. In the evolved 5G and 6G, the Internet of Things (IoT) is used for various purposes and the number of network connected devices is increasing. Typical types of connected devices in a network are not only IoT but also mobile phones and PCs, which may vary each access point or cell. For this reason, better network design and management requires awareness of connected device types based on traffic, service types, and surroundings. Possible approaches for device type classification can be categorized in two classes: analysis based on system logs and management databases; and analysis based on traffic contents and behavior in the network. However, data which network operators can collect is limited to traffic behavior because of privacy law and immature standardization of connected device-related data. Therefore, a new approach to classify device types in the network, whose connected devices are not limited to IoT, based on traffic behavior is necessary. This paper proposes a method to classify device types, such as hubs and cameras by analyzing only traffic behavior, without using traffic contents, in two stages. In the first stage, the proposed method classifies general device types, such as IoT and not IoT, by analyzing packet header statistics using machine learning. Then, in the second stage, the proposed method classifies connected devices that are classified as IoT in the first stage into IoT device types, by analyzing time series traffic behaviors using deep learning. We demonstrate that the proposed method classifies device type by analyzing real traffic dataset and outperforms the existing IoT-only device classification methods in terms of the number of types and the accuracy. The proposed method is suitable for automated network design, management, and control where a number of various devices are connected.