Multi-aspects of emotional electrocardiogram classification in combination with musical stimuli and composite features

Abstract

A new evaluative methodology for classifying emotional electrocardiogram (ECG) was designed based on composite features of wavelet transform and recurrence analysis. To this end, the recurrence dynamics of decomposed ECG were analysed. The ECGs of 20 college students (7 females; and 13 males) were recorded during four emotional states induced by music. Emotion recognition was performed using Fisher, quadratic, and linear perceptron. Moreover, the relevance of the proposed recurrent features has been appraised by means of linear discriminant analysis (LDA), principal component analysis (PCA), Kernel PCA, generalised discriminant analysis (GDA), and Laplacian eigenmaps. The results suggest that LDA outperforms the other techniques. The effect of self-assessment ranks and gender on classification accuracies was also examined. Considering self-assessment scores, higher accuracy rates were achieved. Totally, the maximum rate of 96.15% was attained for women. It seems that the proposed algorithm can open a new horizon in emotion recognition.