source: arxiv machine learning: assessing region-level eeg contributions to cognitive workload prediction
level: research
estimating cognitive workload from eeg is important for safety-critical systems, but it is unclear which scalp regions consistently contribute across tasks and datasets. this study introduces a framework that trains models using features from single anatomically defined regions, then evaluates their predictive performance. four public eeg datasets with varied tasks, hardware, and electrode layouts were used. region importance was measured by a model-agnostic performance metric under both mixed-subject and subject-independent protocols, with results combined via rank aggregation for robustness.
frontal and central regions showed the highest and most consistent contributions across datasets and evaluation settings. temporal and parietal regions were moderately useful, while occipital regions contributed the least. the rank-based aggregation helped stabilize importance scores, reducing variability from individual dataset quirks. these patterns held even when models were tested on unseen subjects, suggesting that frontal and central electrodes carry workload-related signals that generalize across people and tasks.
the findings support using fewer, strategically placed electrodes for practical workload monitoring. by focusing on frontal and central sites, wearable eeg devices could be made simpler and more comfortable without losing much accuracy. the region-level approach also provides a transparent way to compare datasets and guide sensor design. future work could extend this to real-time applications and explore how individual differences affect regional contributions.
why it matters: identifying which eeg regions reliably predict cognitive workload can guide the design of simpler, more practical brain-computer interfaces for real-world monitoring.
source: arxiv machine learning: assessing region-level eeg contributions to cognitive workload prediction