Machine Learning Approaches to Evaluate EEG Correlates of Relaxation Between Supine and Sitting Postures in Eyes-closed Condition.

Background: Brain relaxation is attained using several techniques while sleep remains nature's ultimate remedy. Currently, various machine learning (ML) tools are applied to identify and understand the neural correlates of relaxation from the electroencephalography (EEG) signals. Majority of earlier studies focused on comparing power in the EEG bands during eyes-open and eyes-closed resting state paradigm to train the datasets. However, several Yogic practices are performed using sitting and supine positions. Purpose: This study was aimed to elucidate the relaxation correlates in EEG between supine and sitting position during eyes-closed condition using ML classifiers. Methods: EEG signals were recorded on five different days from O1, OZ, O2, C3, CZ, C4, F3, FZ and F4 brain region using nine unipolar electrodes for 25 minutes during eyes-closed supine and eyes-closed sitting postures each on, along with electrocardiogram (ECG) for heart rate variability (HRV) analysis in a healthy participant. Relaxation was assessed by extracting the relative power of the alpha and theta waves from the EEG data and corroborated with the alpha and theta lateralisation index (LI) and HRV parameters. These EEG metrics were analysed by leveraging ML classifiers (K-nearest neighbours (KNN), support vector machine(SVM), random forest (RF) and XGBoost) for relaxation states under sitting and supine states. Results: Out of all the used classifiers, performance indices of SVM excelled in classifying relaxation states from the EEG alpha and theta band data that was verified with the HRV data and correlated with LI. Conclusion: This study demonstrates that ML especially the SVM was effective in classifying the relaxation states during different postures from the EEG. LI and HRV metrics effectively decoded the underlying message in the EEG and ECG respectively.