Machine learning performance in EEG-based mental workload classification across task types: a systematic review.

Area of Science:

• Neuroscience
• Machine Learning
• Human-Computer Interaction

Background:

• Assessing mental workload (MWL) is crucial for optimizing performance and preventing errors.
• Existing literature lacks standardized tasks for evaluating MWL estimation models, hindering direct comparison of machine learning (ML) approaches.
• Electroencephalography (EEG) is a common modality for non-invasively measuring brain activity related to MWL.

Purpose of the Study:

• To comprehensively examine the performance of ML models in EEG-based MWL classification across different experimental task types.
• To identify how task complexity, specifically single-tasking versus multitasking, influences the accuracy of MWL estimation.
• To provide insights into the current state-of-the-art and identify research gaps in EEG-based MWL estimation.

Main Methods:

• A systematic categorization of existing ML studies based on the task type employed (single-tasking vs. multitasking) and MWL rating method (quantitative vs. subjective).
• Comparative analysis of the performance metrics (e.g., classification accuracy) of ML models across these categories.
• Literature review focusing on EEG-based MWL classification studies.

Main Results:

• A significant decrease in MWL classification accuracy was observed for top-performing ML models in multitasking studies using quantitative task load ratings compared to single-tasking studies.
• Studies employing subjective MWL ratings showed higher classification accuracy than those using quantitative ratings in multitasking scenarios.
• Performance disparities underscore the difficulty of accurately estimating MWL in complex, real-world multitasking environments.

Conclusions:

• Estimating mental workload using EEG-based ML models is significantly more challenging during multitasking, especially when relying on quantitative task load measures.
• The findings highlight the need for developing more robust ML models capable of handling the complexities of real-world multitasking scenarios.
• Further research is needed to establish standardized benchmarks and improve MWL estimation accuracy in dynamic and complex task environments.