Um sistema multimodal para a deteção de stress

• Autor(a) principal: Correia, Hugo André Viana
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.22/21710
• topic: Stress; Machine Learning; Classification; Heart Rate Variability; Facial Expressions; Eye Blink; Pupil Diameter; PERCLOS

Stress is the physiological or psychological response to internal or external factors, which can happen in short or long terms. Prolonged stress can be harmful since it affects the body, negatively, in several ways, thus contributing to mental and physical health problems. Although stress is not simple to properly identify, there are several studied approaches that solidify the existence of a correlation between stress and perceivable human features. In order to detect stress, there are several approaches that can be taken into consideration. However, this task is more difficult in uncontrolled environments and where non-invasive methods are required. Heart Rate Variability (HRV), facial expressions, eye blinks, pupil diameter and PERCLOS (percentage of eye closure) consist in non-invasive approaches, proved capable to accurately identify the mental stress present in people. For this project, the users’ physiological signals were collected by an external video-based application, in a non-invasive way. Moreover, data from a brief questionnaire was also used to complement the physiological data. After the proposed solution was implemented and tested, it was concluded that the best algorithm for stress detection was the random forest classifier, which managed to obtain a final result of 84.04% accuracy, with 94.89% recall and 87.88% f1 score. This solution uses HRV data, facial expressions, PERCLOS and some personal characteristics of the user