Physiological signal-based drowsiness detection using machine learning: Singular and hybrid signal approaches.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Driver drowsiness is a major cause of global road accidents and fatalities.
• Current driver drowsiness detection systems often lack sufficient reliability and validity due to reliance on singular metrics.

Purpose of the Study:

• To evaluate the effectiveness of singular versus hybrid approaches for driver drowsiness detection.
• To investigate the utility of physiological signals (EEG, EOG, ECG) and subjective sleepiness scales in detecting drowsiness.

Main Methods:

• Collected physiological data (EEG, EOG, ECG) during a psychomotor vigilance test (PVT).
• Extracted key features from physiological signals and used Karolinska Sleepiness Scale for ground truth.
• Developed and compared four supervised machine learning models for drowsiness detection.

Main Results:

• Singular physiological measures exhibited performance trade-offs (e.g., high sensitivity, low specificity).
• Hybrid biosignal-based models demonstrated a more balanced performance profile, reducing metric disparities.

Conclusions:

• Hybrid approaches utilizing selected physiological features yield superior performance compared to singular methods.
• The findings support the use of hybrid systems for enhanced in-vehicle driver drowsiness detection, considering practical application factors.