Statistical validation on a larger and more diverse dataset

Establish the statistical significance of the ESP32-S3 binary sleep–wake classification system’s evaluation using a larger and more diverse dataset.

Background

The deployed system was functionally tested in car and bicycle settings with two participants and ten 30-second scenarios per condition. These tests showed reliable wake detection and generally successful detection of pronounced sleep-related head poses, but they do not provide a sufficiently large or diverse basis for statistically significant performance assessment.

A larger and more diverse dataset is therefore needed to assess whether the motion-based model, camera-based pose classifier, and hierarchical sensor-fusion logic generalize across participants, environments, motion conditions, and sleep-related behaviors. The paper explicitly defers this evaluation to future work.