Predicting Stress in Teens from Wearable Device Data Using Machine Learning Methods

Stress management is a pervasive issue in the modern high schoolers life. Despite many efforts to support adolescents mental well-being, teenagers often fail to recognize signs of high stress and anxiety until their emotions have escalated. Being able to identify early signs of these intense emotional states and predict their onset using physiological signals collected passively in real-time could help teenagers improve their awareness of their emotional well-being and take a more proactive approach to managing their emotions. To evaluate the potential of this approach, we collected data from high schoolers with Empatica E4 wearable health monitors (wristband) while they were living their daily lives. The data consisted of stressful event reports and physiological markers over the course of 4 weeks. We developed a random forest model and a support vector machine model and systematically assessed their performance in terms of predicting the onset of stress events and identifying physiological signals of stress. The models showed strong performance in terms of these measures and provided insights on physiological indicators of adolescent stress.