ECG based quantification and modeling of physiological reactions to emotional stimuli

Emotion recognition systems are designed to help in the identification of human emotions, being associated with learning and decision-making on daily tasks as well as treatment and diagnosis in mental health contexts. The research in this area explores different aspects ranging from the information conveyed in different physiological signals to different methods aiming feature selection and emotion classification. This work implements a dedicated experimental protocol to acquire physiological data, such as the electrocardiogram (ECG), while the participants watched videos associated with emotional stimulation to provoke reactions of fear, happiness, and neutral. Data analysis was based on ECG features, being clear that the intended stimuli effectively provoked variation in the heart rhythm and in other ECG features. In addition, each emotional stimulus presents different degrees of reactions clearly distinguished by a clustering procedure. A machine learning model developed based on Support Vector Machine achieved accuracy above 87.01% (training) and 38.40% (test). The emotion state identification was performed over the goals of this study, indicating the potential ability of electrophysiological signal processing for automatic emotion stratification, after emotional video stimulation.