Machine learning techniques for detection of brain state associated with Himalayan Yoga meditation

Stress is a significant risk factor for various mental disorders and, hence, affects the quality of life. In order to live a stress-free lifestyle, meditation has been practiced since the early eras. The purpose of this work is to design a classification model for EEG signals based on Himalayan Yoga meditation and breath-focused control condition by using various Machine Learning (ML) and Deep Learning (DL) techniques. Subject-independent evaluation using Group 10-fold cross-validation approach has been carried out, ensuring subject-wise separation between training and evaluation data. Four conventional ML classifiers: Logistic Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGB), Support Vector Machine (SVM), and three DL classifiers: Inception Convolutional Recurrent Neural Network (IC-RNN), Deep Gated Recurrent Neural Network (DGRNN), and Convolutional Neural Network (CNN) were designed and compared. An Average accuracy of 50.27% was achieved for LR, 74.18% for XGB, 65% for RF, and 54.35% for SVM. An average accuracy of 83.40% was attained for DGRNN, 93.53% was achieved for IC-RNN and 55.52% was attained for CNN with Group 10-fold cross-validation. Moreover, comprehensive feature-level study has been completed to highlight the impact of Himalayan Yoga meditation and breath-focused control condition on human brain activity.