LNNIoT: A Liquid Neural Network-based Bit-level Method for Classifying IoT Devices

The rapid growth of the Internet of Things (IoT) has changed many industries by connecting everyday objects and industrial devices with advanced technology. Therefore, monitoring IoT network traffic is crucial for ensuring the quality of services, enhancing security, and managing traffic control. IoT environments constantly change, with devices frequently added or removed to meet new needs. These devices use different communication protocols like HTTPs, MQTT, and CoAP, making it difficult to classify network traffic accurately. Performing IoT traffic classification under limited-communication scenarios (i.e. Remote Sensors, Wearable Devices, IndustrialIoT, etc.) is essential due to the rapid growth of IoT devices and the need to save computational costs. However, traditional Machine Learning (ML) and Deep Learning (DL) methods for IoT traffic classification face challenges: ML methods require frequent retraining with new data, and DL methods need large amounts of data and are time-consuming. To address these issues, we propose LNNIoT, a robust Liquid Neural Network-based bitlevel method, for classifying IoT devices. LNNIoT is ideal for scenarios where IoT devices are mobile; thus, minimal data is captured for training. LNNIoT learns efficiently from small training sets without retraining. It extracts only initial n-bits from flow payloads of IoT network traces for fast and accurate device classification. Experiments on a public dataset showed LNNIoT achieving average recall of $\mathbf{9 9. 8 2 \%}$ with 128 bits.