Cognitive Workload Classification Using Electroencephalogram Signal

Cognitive workload classification through EEG signals is an emerging field, offering objective, non-invasive insights into mental stress that influences cognitive performance and health. Traditional domain-specific feature extraction methods require extensive expertise and are constrained by manual tuning, while deep learning techniques, though generalizable, require high computational capacity and lack interpretability. To address these issues, we propose a tunable, multi-resolution feature extraction framework that integrates Wavelet Packet Decomposition (WPD) and Common Spatial Pattern (CSP) analysis. This framework allows multiple optimization, including the selection of frequency bands, desired bandwidth, and wavelet functions, offering a customizable approach for improved generalization and efficiency. With an ANN-based classifier our method achieves an accuracy of 94.76%, precision of 94.76%, recall of 94.30%, and an F1 score of 95.21% on the SAM40 dataset, and an accuracy of 98.41%, precision of 98.97%, recall of 97.48%, and an F1 score of 98.22% on the EEGMAT dataset. Our analysis also highlights the role of low-frequency bands in improving cognitive workload classification performance. This study proposes a streamlined and interpretable framework with significant potential for stress diagnosis applications.