Dry EEG-based Mental Workload Prediction for Aviation

Human errors are reputed to be one of the main causes of most accidents or incidents in aviation. Such can be explained by the fact that pilots are frequently exposed to sources of emotional and cognitive stressors, including challenges pertaining to mental workload. Real-time mental workload assessment of crew during flight could help identifying cognitive overload of the crew and then reduce aviation accidents. Electroencephalography (EEG) is a well-known tool used to infer mental states. EEG prediction of mental workload is however mostly performed in highly controlled settings whereas, in the cockpit, many types of confounds (e.g., physical movement) can induce noise into the signal. The goal of this study was to develop an EEG-based workload prediction model that could be used in aviation use cases characterized by noisy signal. To reach this goal, we used machine-learning algorithms to explore the feasibility to classify different levels of mental workload from EEG features. We used a dataset composed of noise induced by physical activity (either low, medium or high levels) collected while participants performed either low-demanding or high-demanding cognitive tasks. Using only three electrodes (Fp1, Fp2 and P3) and 15 spectral band features with the physical condition label, we generated a random forest classifier with a prediction accuracy of 76%. This model could run in real time and provide workload inference at a rate of one prediction each second. Overall, our results show the possibility for predicting real-time mental workload in operational environments using dry-electrode EEG solutions developed for ambulatory use cases.