Assessment of flight fatigue using heart rate variability and machine learning approaches.

The accurate identification of flight fatigue is crucial for managing pilot training intensity and preventing aviation accidents. However, as a subjective perception, flight fatigue is often difficult to evaluate objectively. Heart rate variability (HRV), derived from electrocardiogram signals and regulated by the autonomic nervous system, is recognized as an effective biomarker for assessing fatigue status. This study proposes a novel HRV-based method for the automatic and objective classification of flight fatigue. This study involved an experimental investigation conducted with a cohort of 90 pilots. First, we conducted statistical analyses to investigate whether HRV features and respiratory rate indicators significantly differed across various fatigue levels. A subset of HRV features and the respiratory metric were used as input variables for four machine learning algorithms: decision tree, support vector machine, K-nearest neighbor, and light gradient-boosting machine (LightGBM). These models were applied to perform a three-level classification of flight fatigue. Finally, classification performance was evaluated using average accuracy, precision, recall, and F1 score. Among these models, LightGBM demonstrated the best performance, achieving an accuracy of 0.886 ± 0.057, precision of 0.837 ± 0.064, recall of 0.861 ± 0.086, and F1 score of 0.849 ± 0.067. These findings indicate that a LightGBM model trained on 12 selected HRV features and one respiratory indicator can accurately categorize flight fatigue into three levels. Fatigue can be detected even when mild, enabling real-time monitoring and early warning of flight fatigue. This approach holds potential for reducing fatigue-related flight accidents.