Cognitive performance detection using entropy-based features and lead-specific approach

Detecting cognitive performance during mental arithmetic allows researchers to observe and identify the brain’s response to stimuli. Existing non-invasive methods for automated cognitive performance detection need improvements in terms of accuracy. In this work, a novel approach for cognitive performance has been proposed which uses short-duration electroencephalography (EEG) signal (4.094 s). Stationary wavelet transform (SWT) has been used to decompose the signal followed by extraction of entropy-based features and classification using selected attributes. To tackle the imbalanced data issue, adaptive synthetic sampling approach has been used. The proposed technique works in two modes: multi-lead approach (MLA), where EEG signal from multiple leads was used, and a novel lead-specific approach (LSA), where EEG signal from a single lead (F4) was used. A high accuracy of 94.00% in MLA and 93.70% in LSA reflects reliability of the proposed technique. The use of short-duration single-lead EEG signal makes this technique suitable for continuous monitoring system of cognitive performance during mental workload.