Emotion identification from physiological signals using iterative filtering-based empirical wavelet transform

Abstract

This study proposes a novel framework for human emotion identification based on multi-modal physiological signals, namely electrocardiogram (ECG), electroencephalogram (EEG), and phonocardiogram (PCG). The use of signal analysis method and machine learning (ML) technique can help in designing an intelligent system in order to identify emotions from the physiological signals. In this work, a new signal analysis method termed as iterative filtering-based empirical wavelet transform (IF-EWT) has been proposed for decomposing the non-stationary physiological signals into simpler components. The features such as dispersion entropy and band power have been computed from the decomposed components and classified into four emotional states, happy, sad, fear, and neutral, using ML classifiers. The studied ML classifiers include quadratic discriminant analysis (QDA), k-nearest neighbor (KNN), support vector machine (SVM), ensemble bagged trees (EBT), and neural network (NN). The proposed framework has been evaluated on multi-modal physiological data recorded during participant's exposure to emotion-eliciting audiovisual clips. The proposed framework achieved an average classification accuracy of 97.17% for the subject-dependent case and 93.6% for the subject-independent case using the KNN classifier. The presented work is this paper has applications in various fields such as mental health monitoring, personalized human-computer interaction, and affective computing. © 2025 Elsevier B.V., All rights reserved.