Towards Practical Deployment: Subject- Independent EEG-Based Mental Workload Classification on Assembly Lines

Abstract

Despite significant advancements in Electroencephalography (EEG)-based Mental Workload (MWL) assessment facilitated by deep learning, challenges such as subjectindependent MWL estimation persist. Addressing this challenge is crucial for the widespread adoption of the technology in practical, real-world settings. It could facilitate the deployment of neuroadaptive systems across various users without the need for individual calibration, significantly reducing setup time and complexity, and enhancing the scalability. This study explores subject-independent MWL estimation under realistic conditions of a typical assembly line workplace, as opposed to the idealized settings typical of existing research. We employed a convolutional neural network (CNN) to classify 10s EEG segments into two MWL categories, based on different complexity of visual instructions for manual assembly. The results in subject-dependent and subject-independent cases were compared. The findings reveal only a marginal decrease in classification accuracy when transitioning from subject-dependent (92.2%) to subject-independent scenarios (90.8%). The study demonstrates the feasibility of using deep learning models for EEG-based MWL estimation under realistic conditions, paving the way for broader applications of this technology across diverse industrial environments.