Classification of Native and non-native songs, neutral music, and resting state of human brain from EEG-based auditory evoked potential using explainable features

Document Type

Article

Publication Date

7-14-2026

Abstract

Auditory evoked potentials (AEPs) provide a non-invasive method for studying auditory processing, with uses in cognitive monitoring and biomedical applications. Although there have been studies of music-based EEG paradigms, a clear connection between AEP features and interpretable neurophysiological mechanisms that differentiate complex auditory categories, such as familiarity and linguistic content, remains lacking. To address this gap, this study proposes an interpretable EEG-based framework for classifying native songs, non-native songs, neutral music, and resting states using physiologically meaningful features extracted from electroencephalographic signals, including time and time-frequency domain representation. The interpretability of the proposed framework is achieved at the feature level, where selected features and their importance are directly associated with known brain rhythms and cortical functions. These auditory categories facilitated the study in examining how familiarity, novelty, and emotional engagement distinctly influenced auditory evoked potentials. The findings indicate that delta brain rhythms can serve as markers for identifying AEPs to auditory stimuli, an exploration conducted for the first time to our knowledge. The extracted features show distinct cortical engagement patterns during music-evoked activity. Using this, we reliably distinguish native songs, non-native songs, neutral music, and resting-state activity, with an average accuracy of 97.5% using a random forest classifier. Empirical mode decomposition further reveals intrinsic mode functions that enable precise identification of the P300 component within auditory evoked responses. The framework achieved 76.36% average accuracy across all binary LOSO comparisons with SVM, peaking at 84.86% for native vs. resting tasks. The proposed framework can monitor delta-band cortical responses during music listening, which can aid meditation, neurofeedback, and rehabilitation by tracking auditory engagement and relaxation.

Publication Source (Journal or Book title)

Neurocomputing

This document is currently not available here.

Share

COinS