Robust multimodal mental workload classification: A cross-physiological condition machine learning approach.

Area of Science:

• Aviation Physiology
• Machine Learning in Healthcare
• Human Factors Engineering

Background:

• Pilots face high mental workload (MW) exacerbated by hypoxia and sleep restriction.
• Existing MW predictive models may lack robustness across varying physiological states.

Purpose of the Study:

• To cross-validate machine learning MW predictive models under hypoxia and sleep restriction.
• To develop a robust MW predictive model using multimodal physiological data for improved validity.

Main Methods:

• Seventeen participants underwent controlled sleep restriction (SR) or habitual sleep (HS) and hypoxia (HY) or normoxia (NO).
• Mental workload was manipulated using the Multi-Attribute Test Battery (MATB)-II and an auditory task.
• Machine learning classifiers were evaluated using features from EEG, ECG, respiratory, and eye-tracking sensors.

Main Results:

• Individual models performed best under habitual sleep and normoxia (e.g., KNN F1 score 80.3%).
• Model performance significantly decreased under sleep restriction and/or hypoxia (F1 <35%).
• Models trained on data from all conditions, especially those incorporating EEG and eye-tracking, showed improved cross-condition performance (e.g., KNN F1 score 77.4%).

Conclusions:

• Training machine learning models on diverse physiological conditions is crucial for robust MW prediction.
• Multimodal physiological data enhances the validity and reliability of MW predictive models in aviation.