Survey and Experimentation to Compare IoT Device Model Identification Methods

The widespread use of the Internet of Things (IoT) devices introduces a novel security threat due to the many vulnerable devices connected to the Internet. One effective countermeasure against such threats involves detecting vulnerable IoT devices by identifying the models of connected IoT devices. Several studies have concentrated on model identification methods that utilize flow and communication data to identify IoT device models. With the advances in machine learning and deep learning, high identification accuracy has been reported under certain conditions. However, when implementing these research findings, selecting the most suitable model for the application is crucial. This selection process presents challenges in terms of reproducibility and applicability. The issue of reproducibility lies in the difficulty of implementing a model that accurately reproduces the methods outlined in the paper. The applicability issue arises when attempting to select the best method based solely on the experimental results described in the papers, as difference studies have conducted these experiments on different datasets across various papers. This study focused on IoT device model identification using flow data. We surveyed existing studies and selected a method with high reproducibility. Additionally, we conducted experiments to evaluate the accuracy of the selected methods using multiple datasets under uniform conditions for feature values and dataset usage. Consequently, we identified the optimal method in terms of both reproducibility and applicability within the scope of our study.