Can mental workload in EEG tasks be classified using machine learning algorithms?

Abstract

Measurement of mental workload is an important problem in human computer interaction (HCI) circles. With the rapid development of technology in the field of HCI, there is a need for creating strategies to automate the measurement of mental workload. One approach would be to classify mental workload levels from physiological measurements, such as electroencephalogram (EEG). Accordingly, the purpose of this study was to classify mental workloads by using machine learning algorithms based on EEG signals. To this end, this study used an existing dataset for users who had their EEG signals collected while performing a cognitively demanding task. After the task, users also provided subjective assessments of the mental workload they experienced, which were used to train a supervised learning model. The results indicate that several types of machine learning classifiers, including k-Nearest Neighbors (kNN), Decision Tree (DT), and Random Forest (RF), can effectively identify low, medium and high workload levels. kNN (89.23%) and RF (90.76%) are the most effective among the algorithms according to their F1 score. These results indicate that mental workload can be classified based on wireless EEG brainwaves. Results also indicate that EEG brainwaves should be standardized instead of being normalized to the range [0, 1]. In the future, deep learning algorithms should be explored for their potential enhance mental workload classification in EEG tasks.