Meditation-induced cognitive state classification using sparse brain connectivity features and electroencephalography
Intelligent Data Analysis September 17, 2025 Soniya Usgaonkar, Damodar Reddy Edla, R Ravinder Reddy
A brain-computer interface (BCI) approach to meditation uses a multivariate auto-regressive (MVAR) model of EEG signals that captures both temporal dynamics and interactions between electrodes. Sparsity is introduced via the group least absolute shrinkage and selection operator (GLASSO) to reduce volume conduction. From the sparse coefficient matrix, connectivity features like average energy value, phase lag value, mean absolute correlation, and magnitude squared coherence are extracted. These features help classify EEG into three meditation states: EM, NM, and CO. Among several classifiers, decision trees performed best with 97.12% accuracy, 96.12% precision, 97.39% recall, and an F1-score of 97.01%. Adding sparsity to the MVAR model improves EEG classification of meditation states.