English

Abstract

This study is related with emotion recognition based on autonomic nervous system responses. Three different emotional states, fear, surprise and stress, are evoked by stimuli and the autonomic nervous system responses for the induced emotions are measured as physiological signals such as skin temperature, electrodermal activity, electrocardiogram, and photoplethysmography. Twenty-eight features are analysed and extracted from these signals. The results of one-way ANOVA toward each parameter, there are significant differences among three emotions in some features. Therefore we select eight features from 28 features for emotion recognition. The comparative results of emotion recognition are discussed in view point of feature space with the selected features. For emotion recognition, we use four machine learning algorithms, namely, linear discriminant analysis, classification and regression tree, self-organizing map and naive bayes, and those are evaluated by only training, 10-fold cross-validation and repeated random sub-sampling validation. This can be helpful to provide the basis for the emotion recognition technique in human computer interaction as well as contribute to the standardization in emotion-specific ANS responses.