Real-time EEG-based prediction of mental workload in a multitasking environment

In evaluating new Air Traffic Control (ATC) automation, it is critical to objectively measure its impact on controllers' mental workload. To this end, we investigated neurophysiological correlates of traffic load in ATC simulations using electroencephalography (EEG). We had three goals: 1) develop machine learning methods that predict mental workload imposed by varying traffic levels (taskload), 2) identify EEG features that correlate to taskload, and 3) understand individual differences in EEG features that measure workload. In this study, eight subjects matched for gender and age performed realistic 'human-in-the-loop' ATC simulations with task-load as a within-subject variable. This multitasking environment required controllers to intermix visual, speaking, motor, and listening tasks. Our Gaussian process regressor achieved moderate power in predicting task-load. Comparisons of significant features between linear regression analysis and machine learning indicate that beta and gamma were consistently significant and were among the most informative EEG features for all subjects. These results are consistent with studies using the MATB and military training exercise, suggesting that beta and gamma power increases are related to workload in multitasking environments. Gamma effects were more global, whereas beta effects were occipital-temporal, distinct from the frontal, central and parietal taskload effects reported by others in their ATC task. However, their ATC task did not include speaking and listening, which could explain why our subjects had more temporal effects. These results indicate that EEG-based measures of workload can be developed for multitasking environments such as ATC.