Full Program »
Smartphones and Wristbands Detect Stress As Good As Intrusive Physiological Devices
Detecting stress timely is relevant to help people's mental and general health. Multiple researchers explored different aspects of this problem by analysing physiological responses of the sympathetic nervous system (changes in heart rate, breathing rate variability, skin temperature or conductance) or by studying the person's activity levels throughout the day by taking data from their smartphone use. Currently, smart devices can provide relevant information associated with stress depending on their usage. Combined with machine learning systems, it is possible to obtain high accuracy levels in detecting stress, some reaching up to 90\% under ideal conditions. This study aims to detect stress at different intrusion levels (low for smartphones and medium or high for devices recording physiological data) to focus on reducing intrusiveness and keeping a reasonable accuracy level. Our stress detection models obtain up to 75% in the wild for low intrusion and 97% for medium or high intrusion levels by using public datasets. Moreover, high intrusion devices do not improve the quality significantly with respect of the medium intrusion ones (between 1 and 3 points). These promising results show that, by using only smartphones and wristbands, we can obtain confident detection, similar to invasive methods, by providing a non-intrusive procedure to help millions of people to detect and deal with stress.