Discrete to dimensional physiological emotion classification

• Autor(a) principal: Alves, Carolina Fernandes
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/33692
• topic: Emotion; Emotion elicitation; Physiological signals; Classification; Discrete model; Dimensional model

Emotions play a very important role in human life. The way we communicate and interact with others, our actions, thoughts, are all influenced by them, whether in a positive or negative way. Unfortunately, there is a variety of mental diseases, like anxiety and depression, that are characterized by a prevalence of negative emotions, and in which people tend to have a higher difficulty in understanding their emotional state. Consequently, the importance of each one of us being able to identify their emotional state is crucial to guarantee a healthy control over emotions. The emotion recognition systems can be one of the solutions to help people identify and interpret their emotions, hence increasing their well-being and health. Studies in this area have explored different topics ranging from the type of signals and features to the method of feature selection and emotional classification. Furthermore, they also began to diverge in the approach of describing emotions, which can be discrete or dimensional. In this work, the two approaches were studied to understand the impact of the emotional description on the classification process and to conclude on the most adequate approach to identify emotions. To this end, it was created a database of the physiological signals: electrocardiogram, electrodermal activity and electromyogram of the medial frontalis and trapezius muscles. An exploratory analysis was performed with these data revealing that the electrocardiogram and electrodermal activity represent the most informative in emotion discrimination. Nevertheless, in a multivariable approach, the features from electromyogram reveal to be useful on emotion classification. The approach initially studied was based on a discrete model of emotions, however, misclassification of some observations led to considering the hypothesis of testing a dimensional model (valence/arousal). This model proved to be more robust than the previous one, which led to the conclusion that it is better adapted to both emotional response and the individual response to the stimulus, confirming its best description of the emotion.