A Machine Learning Approach to Reduce Mental Fatigue Risk of Pilots Based on HRV Data

Fatigue is important in aviation because the risks associated with fatigue pose a threat to many people. The purpose of this study is to investigate the possibility of using a wearable ECG device to detect mental fatigue in pilots. 22 healthy individuals were recruited from Beihang University and each subject wore a Polar H10 heart rate band with a Polar IGNITE sports watch to record heart rate variability (HRV) data. To stimulate the fatigue state, each subject performed a simulated flight experiment of 1h and the number of blinks was recorded by the face recognition method as a label for fatigue. HRV data and blink frequency were collected every 5 minutes throughout the experiment. After feature selection, four HRV indicators SDNN, LF_HF, LF and VLF were extracted for subsequent analysis. Five machine learning algorithms: support vector machine (SVM), Random Forest, AdaBoost, K-nearest neighbor (KNN), Naïve Bayes (NB) were used to build a classifier for automatic detection of pilot fatigue. Naïve Bayes (NB) had the best performance and had an average CV F1 score of 0.9065 on the training set and 0.9250 on the test set.