Estimating vigilance from the pre-work shift sleep using an under-mattress sleep sensor.

Area of Science:

• Occupational Health
• Sleep Science
• Machine Learning

Background:

• Vigilance impairment in shift work increases workplace errors.
• Current fatigue prediction relies on multi-night, manually entered sleep data.
• Developing efficient, accurate fatigue prediction is crucial for worker safety.

Purpose of the Study:

• To develop and validate a machine learning model for predicting vigilance errors.
• To utilize data from a single sleep period captured by an under-mattress sensor.
• To assess the model's accuracy against established fatigue prediction methods.

Main Methods:

• Employed extra-trees machine learning models to predict psychomotor vigilance task (PVT) performance.
• Utilized sleep data from an under-mattress sensor following a single sleep period.
• Compared model predictions (reaction time, speed, lapses) with actual PVT performance during simulated night shifts.

Main Results:

• The model achieved moderate accuracy in predicting PVT performance metrics.
• Standard errors for reaction time, speed, and lapses were 19.9 ms, 0.42 reactions/s, and 7.2, respectively.
• The model correctly classified 84% of trials with significant lapses (Matthews correlation coefficient = 0.59).

Conclusions:

• A machine learning model can predict vigilance errors from a single night's sleep data.
• This sensor-based approach offers comparable performance to current, more cumbersome methods.
• The findings suggest a potential for improved fatigue management and safety in shift work.