Stress causes a significant reduction in quality of life, with 37% of adults globally reporting substantial daily stress in 2024. Work is a leading cause of stress. 69% of Americans listed their occupation as a significant source of stress, and more than half of Japanese workers reported extreme work stress, with stress associated with negative effects on productivity and work engagement. 

Wearable devices allow us to continuously and passively record elements of our physiology (e.g., electrodermal activity, skin temperature) as we go about our daily routines. We study how daily patterns in these physiological measures relate to different factors of our wellbeing (e.g., stress).

We analyse a large wearable electrodermal activity dataset provided by 381 office workers, containing 247,942 total hours of signal and 1,693 self-reported stress measures. There are two parts to this project. First, we probe significant associations between objective electrodermal activity properties and subjective perceived stress to provide insights on potential wearable biomarkers for stress. Second, we  perform multimodal machine learning analysis to test if stress can be estimated using only passively-sensed wearable features. 

Joint work with NEC and Terumi Umematsu.