Feature Representation for Meditation State Classification in EEG Signal

Meditation in walking posture is less studied than sitting meditation. To distinguish between sitting and walking meditation states, a classification framework was developed using EEG signal features and a random forest classifier. Single-modal features—original power, power ratio, and non-linear dynamics—were compared. The original power feature outperformed non-linear dynamics. A multi-modal feature combining all single-modal features achieved the highest accuracy in identifying the two meditation states.

Study at a glance

Abstract

Meditation has been shown as an efficient way to promote human well-being. Most studies focused on meditation in sitting posture. However, meditation in walking posture was rarely studied. In order to identify these two meditation states (i.e., sitting and walking), we proposed a classification framework by leveraging different features extracted from the EEG signals and the random forest classifier. This study first investigated different single-modal features, including original power, power ratio, and non-linear dynamics. Further, we also concatenated all the single-modal features into a multi-modal feature. The experimental results show that the original power feature is better than the non-linear dynamics feature in meditation state classification. Moreover, the multi-modal feature outperforms all the single-modal features and can identify sitting and walking meditation with high accuracy.