Reliability over time of EEG-based mental workload evaluation during Air Traffic Management (ATM) tasks.

Machine-learning approaches for mental workload (MW) estimation by using the user brain activity went through a rapid expansion in the last decades. In fact, these techniques allow now to measure the MW with a high time resolution (e.g. few seconds). Despite such advancements, one of the outstanding problems of these techniques regards their ability to maintain a high reliability over time (e.g. high accuracy of classification even across consecutive days) without performing any recalibration procedure. Such characteristic will be highly desirable in real world applications, in which human operators could use such approach without undergo a daily training of the device. In this work, we reported that if a simple classifier is calibrated by using a low number of brain spectral features, between those ones strictly related to the MW (i.e. Frontal and Occipital Theta and Parietal Alpha rhythms), those features will make the classifier performance stable over time. In other words, the discrimination accuracy achieved by the classifier will not degrade significantly across different days (i.e. until one week). The methodology has been tested on twelve Air Traffic Controls (ATCOs) trainees while performing different Air Traffic Management (ATM) scenarios under three different difficulty levels.