A Psychophysiological Model Based on Machine Learning Algorithms for Evaluating Commercial Airline Pilots' Mental Workload in Flight-Simulation Context

Abstract

Pilot errors account for the majority of flight accidents, many of which are influenced by mental workload. This study introduces a data-driven psychophysiological model, utilizing machine learning algorithms, to evaluate pilots' mental workload. We conducted a flight-simulation experiment involving twenty commercial airline pilots, assessing their mental workload under varying levels of task demand and visibility conditions. Psychophysiological responses were recorded, and machine learning algorithms were employed to analyze the data. To evaluate model performance, we used a leave-one-subject-out cross-validation method and calculated area under the curve values. Our findings indicate that psychophysiological metrics vary in their sensitivity to changes in pilots' mental workload. Notably, the Gradient Boosting Decision Tree algorithm demonstrated the highest classification performance under high-visibility conditions, while the Gaussian Naive Bayes algorithm excelled under low-visibility conditions. These results suggest pilots' mental workload can be effectively identified through psychophysiological metrics combined with machine learning algorithms. Furthermore, visibility conditions may influence the model's classification performance. This model offers a complementary approach to the subjective evaluation currently used by flight instructors to assess pilots' mental workload management capabilities during flight training and certification. It also provides a data-driven tool aligned with evidence-based training principles, enhancing the evaluation of pilots' mental workload management capabilities in flight scenarios.