Cognitive load detection through EEG lead wise feature optimization and ensemble classification.

Cognitive load stimulates neural activity, essential for understanding the brain's response to stress-inducing stimuli or mental strain. This study examines the feasibility of evaluating cognitive load by extracting, selection, and classifying features from electroencephalogram (EEG) signals. We employed robust local mean decomposition (R-LMD) to decompose EEG data from each channel, recorded over a four-second period, into five modes. The binary arithmetic optimization (BAO) algorithm employed to reduce the feature space and extract multi-domain features from modes, thereby optimizing classification performance. Using six optimized machine learning (ML) classifiers, we conducted an exhaustive study that encompassed both lead-wise and overall feature classification. We improved our method by combining R-LMD-based multi-domain features with BAO and optimized ensemble learning (OEL) classifiers. It was 97.4% accuracy (AC) at finding cognitive load in the MAT (mental arithmetic task) dataset and 96.1% AC at finding it in the STEW (simultaneous workload) dataset. In the same vein, this work introduces lead-wise cognitive load detection, which offers both temporal and spatial information regarding brain activity during cognitive tasks. We analyzed the 19 and 14 leads for the MAT and STEW, respectively. The F3 lead was notably noteworthy in its ability to analyze a variety of cognitive tasks, obtaining the maximum classification AC of 94.5% and 94%, respectively. Our approach (R-LMD+BAO+OEL) outperformed existing state-of-the-art techniques in cognitive load detection.