Temporal-domain Analysis of Meditation and Mind-wandering EEG Signals for Different Meditation Traditions

Features extracted from the time domain of EEG signals, such as sample entropy and zero-crossing rate, can distinguish meditation from mind-wandering with 93% to 97% accuracy across three traditions: Himalayan Yoga, Isha Shoonya, and Vipassana. Signal-complexity features were most effective for differentiation. The analysis used publicly available EEG recordings, preprocessed to remove noise, and classified with Support Vector Machine and Random Forest algorithms.

Study at a glance

Abstract

Meditation is an age-old practice that continues to attract significant interest in the field of neuroscience due to its stress-reducing and therapeutic benefits. In line with this growing area of research, our study aims to contribute by identifying potential neural markers associated with meditation. To achieve this goal, we investigate various features extracted from the time domain representation of EEG signals during meditation across different meditation traditions, including Himalayan Yoga meditation, Isha Shoonya meditation, and Vipassana meditation, and analyze them. Further, we use these features to discriminate EEG signals recorded during meditation from those during mind-wandering states. To facilitate our exploration, we use EEG signals from an online repository and preprocess them to eliminate noise and artifacts. Next, we extract an array of linear and non-linear features such as mean, standard deviation, RMS, kurtosis, skewness, zero-crossing rate, and sample entropy, which serve as input for two powerful binary classification algorithms: Support Vector Machine and Random Forest. We achieve classification accuracies ranging from 93% to 97% for different groups, indicating the efficacy of utilizing temporal- domain features of EEG signals for the successful differentiation of tasks. Furthermore, we conduct an analysis to identify the most crucial features, particularly their potential as neural markers. Our findings suggest that features associated with signal complexity demonstrate the highest efficacy in distinguishing between conditions and groups.