Enhanced EEG-based cognitive workload detection using RADWT and machine learning.

Understanding cognitive workload improves learning performance and provides insights into human cognitive processes. Estimating cognitive workload finds practical applications in adaptive learning systems, brain-computer interfaces, and cognitive monitoring. In this work, different levels of cognitive workload are investigated, and a classification approach based on the Rational-Dilation Wavelet Transform (RADWT) is proposed. RADWT excels at capturing the oscillatory behavior of EEG signal sub-bands, offering high precision through its ability to adaptively analyze both temporal and spectral dynamics. Different classifications of machine learning and feature selection techniques were evaluated to get optimum classification accuracy and identify the most effective combination of features for the used dataset. The analysis shows that the most relevant brain region in differentiating cognitive workload levels is the frontal region, along with alpha and theta rhythm sub-bands. Integrating RADWT with a Linear Support Vector Machine (LSVM) and minimum Redundancy Maximum Relevance (mRMR) feature selection method yields notable classification accuracy. Concretely, the model yields accuracies of 96.6% for 0-back vs.3-back, 94.9% for 0-back vs 2-back, 92.3% for 2-back vs 3-back, and 81.7% for the three-class scenario. These results confirm the validity of the method proposed for estimating cognitive workload using the RADWT- and machine learning-based approach. The results also offer insights into neural mechanisms and a foundation for advanced applications in adaptive systems, brain-computer interfaces, and cognitive monitoring.