The Effect of EEG Segment's Length on Mental Workload Detection

The rapid advancements in neurotechnology and neuroergonomics have propelled the field of passive brain-computer interfaces (pBCIs) to the forefront of human-computer interaction. One critical aspect of pBCIs is the detection of mental workload (MWL), which plays a pivotal role in optimising cognitive performance and preventing cognitive overload. This study looks into the best length of an electroencephalography (EEG) segment for detecting mental workload (MWL) with passive brain-computer interfaces (pBCIs). We used the STEW dataset, which has EEG data from 48 people doing different cognitive tasks, to look at three different feature sets and five machine learning models to see how segment duration affected the accuracy of detection. We observed a rapid initial increase in accuracy with increasing segment duration, followed by a plateau that resulted in minimal improvement. Using gradient analysis and the Kneedle algorithm, the best segment length was found to be 2.43 seconds on average. This was the best balance between information content and computational efficiency. These results go against what most people think and give us a way to measure how to improve MWL detection in pBCIs, which could lead to better interactions between people and computers. We need to conduct further research to investigate the generalizability of these findings to different datasets and task paradigms, and to examine the influence of specific feature sets and classification algorithms on the ideal segment duration for MWL detection.