DIFORMER: An Attention-Based IoT Device Identification Method

The Internet of Things (IoT) has significantly expanded the scope of the Internet by enabling seamless communication between devices and the cloud. However, the rapid proliferation of IoT devices has led to a surge in cyber-attacks, such as Distributed Denial of Service (DDoS) attacks, Man-in-the-Middle (MITM) attacks, and privacy data theft. These threats underscore the urgent need for robust device identification (DI) methods. However, current DI approaches often struggle with issues related to accuracy, generalizability across different network environments, privacy concerns, and real-time performance. To address these issues, we introduce DIFORMER, a novel method that leverages network packet data and advanced deep learning techniques, specifically the attention mechanism and residual networks. DIFORMER outperforms existing methods, even in privacy-limited scenarios, and is more generalizable as it does not strictly rely on features like MAC and IP addresses. Rigorous experiments confirm DIFORMER’s effectiveness and robustness, making it a promising solution for securing IoT environments.