Predicting Pilot Workloads Using Physiological Measures

Abstract

In this study, we utilized machine learning algorithms to predict pilot workload based on physiological, cognitive, and eye-tracking data collected from 7 pilots performing tasks in an unclassified F-35 flight simulator. We used the Bedford Workload Rating Scale (BWRS) to validate the high workload conditions induced during the simulated flight scenarios. We then trained models and compared their performance at predicting high workloads. Our results showed model performance was higher when classifiers were trained on individual pilots instead of on a group of pilots. We found that changes in Percent Change Pupil Size (PCPS), an eye-tracking measurement, were particularly noticeable in high vs. low-workload scenarios. This metric emerged as the most significant factor in distinctly difficult situations. These findings suggest a shift towards personalized machine-learning models for enhancing human-machine interactions in aviation through biometric and PCPS monitoring. Future work should examine a more diverse set of tasks, validated by study subjects, to assess the potential benefits of incorporating artificial intelligence (AI) assistance systems into the cockpit.