Study of Stress Detection and Proposal of Stress-Related Features Using Commercial-off-the-shelf Wrist Wearables

This paper discusses the possibility of detecting personal stress making use of popular wearable devices available in the market. Different instruments found in the literature to measure stress-related features are reviewed, distinguishing between subjective tests and mechanisms supported by the analysis of physiological signals from clinical devices. Taking them as a reference, a solution to estimate stress based on the use of commercial-off-the-shelf wrist wearables and machine learning techniques is described. A mobile app was developed to induce stress in a uniform and systematic way. The app implements well-known stress inducers, such as the Paced Auditory Serial Addition Test, the Stroop Color-Word Interference Test, and a hyperventilation activity. Wearables are used to collect physiological data used to train classifiers that provide estimations on personal stress levels. The solution has been validated in an experiment involving 19 subjects, offering an average accuracy and F-measures close to 0.99 in an individual model and an accuracy and F-measure close to 0.85 in a global 2-level classifier model. Stress can be a worrying problem in different scenarios, such as in educational settings. Thus, the last part of the paper describes the proposal of a set of stress related indicators aimed to support the management of stress over time in such settings.