EEG-based Mental Workload Assessment using a Graph Attention Network

Sustained high mental workload (MWL) experienced by operators in high-pressure jobs can compromise their performance, potentially endangering them as well as others. Using electroencephalograms (EEG) to gauge MWL levels is an approach that has been gaining prominence lately. Graph attention networks (GAT) have previously been used to great effect for traffic forecasting, citation networks, etc. In that context, we propose a GAT-based approach for improving the assessment of MWL using EEG signals. We focus on distinguishing EEGs corresponding to a high MWL from the EEGs corresponding to a low MWL and provide a comparative analysis of different features viz. band power, wavelet features, and autoregressive (AR) parameters. The obtained results show that this approach achieves an average accuracy of up to 95.66%, which is superior to that obtained using conventional multilayer perceptron (MLP) and several other recently used methods.