A Multi-Modal Traffic Classification-Based Device Identification Method

Abstract

Internet of things (IoT) devices are widely used in various fields, with their growing diversity and complexity posing challenges for traditional security measures. Device fingerprint identification can enhance network security and reliability by verifying device features. However, traditional device fingerprint identification methods usually rely on a single mode of traffic characteristics. In the face of changing network environments and diversified device types, it is often difficult to ensure efficient identification performance and robustness. To address the above challenges, this paper proposes a multi-modal traffic classification method for device identification in IoT networks to address the challenges in accuracy and robustness posed by traditional single-modal traffic feature approaches. The method combines various traffic features, such as packet size, transmission interval, flow duration, packet rate, byte rate, and protocol number. It includes four modules: data collection, preprocessing, model training, and fingerprint identification. Network traffic data are collected using deep packet inspection and capture tools, and features are standardized. The bidirectional encoder representations from transformers (BERT) model is applied for sensitive text feature extraction, while the convolutional neural network (CNN) model aids in device identification. Experimental results demonstrate high accuracy and robustness across different network environments and device types.