Samenvatting

In this study, eye-tracking and deep learning are employed to classify pilot experience during a twin-engine Piper Seneca flight simulator based on the potential to enhance aviation training and safety through the integration of human cognitive behaviors with artificial intelligence to optimize inter-cognitive mobility in aviation. The goal is to identify data-driven training protocols based on visual behaviors. With Tobii Pro Glasses 3, eye-tracking metrics (gaze distribution, fixation duration, saccade dynamics, pupil diameter) were captured from one experienced and one novice pilot during realistic simulator flights. Metrics were analyzed and input into a TensorFlow-based Keras neural network for binary expertise classification, where SHAP analysis chose significant predictors and visualizations highlighted behavioral differences. The model achieved 98.36% accuracy, ROC-AUC of 0.99, and Matthews Correlation Coefficient of 0.9673. SHAP analysis revealed gaze direction to be the primary discriminator, and experts showed purposeful scans versus novices’ chaotic patterns. Experts had longer fixations (mean: 1465.13 ms) and saccades (mean amplitude: 138.98 px) than novices’ brief fixations (mean: 586.90 ms) and saccades (mean amplitude: 21.87 px), with pupil size changes indicating increased novice cognitive load. The small sample size (two pilots) limits generalizability and requires a larger number of pilot participants for more robustness. Despite this, visual behavior significantly indicates proficiency, supporting eye-tracking-concatenated training systems that deliver instantaneous feedback to enhance situational awareness, reduce mental load, and improve safety. Future studies with larger cohorts and different flight scenarios need to validate and provide additional support for these findings.