Key points are not available for this paper at this time.
Quantifying workload is necessary for effective and personalized flight training of student pilots: their workload must not be too low (risk of boredom) nor too high (overload). Passive brain-computer interfaces (pBCIs) allow for measurement of an individual's workload from their brain activity, however, the performance of pBCIs remains suboptimal due to individual differences and lack of data for classifier training. In this study, we addressed this problem by combining EEG and behavioral data from six novice military pilots who performed a flight task in Virtual Reality in order to develop calibration-free pBCIs for workload assessment. Three pBCI classifiers were trained on EEG spectral power features from theta (4-8 Hz), alpha (8-13 Hz) and beta (13-30 Hz) bands, and an additional behavioral feature derived from pilots' control inputs on the (joy)stick. The models reached average classification accuracies of 0.82, 0.78, and 0.78. The key feature driving the models' performance was EEG theta power from several regions of the brain. The pilots' control inputs (i.e., behavioral feature) did not contribute to the model performance, however, it moderately correlated with several EEG theta power features. The results demonstrate the feasibility of a subject-independent pBCI for calibration-free classification of workload in pilots as well as the importance of theta power at frontal and centro-parietal areas as a metric for real-time monitoring of workload. The use of behavioral control inputs together with fewer but highly predictive EEG features warrants further research.
Weelden et al. (Wed,) studied this question.