Disorder Quantifier Outperforms Entropy in EEG Classification of Motor and Visual States

Feature selection is critical for the performance of Machine Learning (ML) algorithms, as model outcomes rely heavily on input data quality. This study empirically validates two recurrence analysis quantifiers microstate entropy and disorder as discriminative features for EEG-based classification. We analyzed EEG data from individuals across four experimental conditions (resting and cycling, with eyes open or closed) to characterize cortical dynamics in distinct motor and sensory states. Using a Random Forest classifier, we compared the performance of microstate entropy and disorder. Our results demonstrate that cycling significantly reduces cortical complexity, corroborating recent findings. Notably, the disorder quantifier (specifically for microstate size N =4) outperformed entropy, achieving classification accuracies of up to 82% for visual states during movement and 79% during rest. Furthermore, we observed hemispheric asymmetries between parietal and occipital regions and confirmed that integrating data from all eight EEG channels substantially improves classification compared to single-channel analysis. These findings highlight the disorder quantifier as a robust, compact, and interpretable descriptor of cortical dynamics.