Automated Classification of Listening Phases During Mantra Meditation from EEG using empirical Fourier-Bessel Decomposition

Abstract

Mantra meditation (MM) is known to influence cognitive and emotional processes, and electroencephalogram (EEG) has been widely used to study meditation-related neural activity. However, automated identification of distinct auditory listening phases during MM remains challenging due to the non-stationary nature of EEG signals. This shows the need for physiologically interpretable analysis of meditative states while listening mantra. This study utilizes an empirical Fourier–Bessel decomposition (EFBD)-based time–frequency representation (TFR) framework for automated classification of EEG signals recorded during Rudram Mantra listening. Multichannel EEG data from 28 subjects were preprocessed, segmented into fixed-length epochs, and decomposed using EFBD to obtain oscillatory mode components. The Hilbert transform is applied to derive TFR, from which discriminative features are extracted and selected using a minimum redundancy–maximum relevance algorithm. Classification is performed using machine learning models under subject-dependent and leave-one-subject-out validation schemes. The proposed framework achieved an average accuracy of 77.5%, with stable cross-subject performance. Channel-wise and rhythm-wise analyses highlighted consistent phase-dependent variations, particularly in frontal and temporal regions. These results demonstrated that the proposed EFBD-based TFR framework has been proven effective in monitoring meditative states of subject while listening mantra.