Are we there yet in terms of practicality and ML-based IoT device identification for security?
Lightning Talk
In my presentation, I will show how we systematically evaluated ML-based IoT identification systems, leveraging curated datasets and model explainability. Through the presentation, I will show several factors that impact the performance of IoT identification models in network deployments. These factors include variability in device operation modes, spatial, temporal, and portability. For example, when using identical devices in different environments, the performance of IoT identification models declines dramatically. Additionally, the performance of machine learning-based identification models for IoT devices can diminish within as early as two weeks.
