CHS: Medium: Collaborative Research: Managing Stress in the Workplace: Unobtrusive Monitoring and Adaptive Interventions
Grant
Overview
Affiliation
Other
View All
Overview
abstract
Workplace stress is a serious problem that has a direct and negative impact on health, happiness, and productivity. Current approaches for both measuring stress and reducing it are limited; measurements typically rely on self-report or obtrusive sensors, while people often don''t seek treatment until the stress has built to dangerous levels (or at all, if they are afraid of other people''s judgments). Common workplace sources of stress are noise, distractions and time pressure. This project''s goal is to develop methods both to detect stress and provide personalized relaxation exercises, in real time and in the work context. To detect stress, the research team will apply machine learning to study how well data from commonly available devices at work such as webcams, fitness trackers, and keyboards can predict individuals'' stress levels. To reduce stress, the team will develop a suite of brief relaxation exercises and a system that uses predicted stress levels to recommend different exercises, learning over time which ones work best for a particular person. These predictive models and interventions will be tested in a long-term study in a real office environment, both validating the work and providing direct effects on experimental participants'' well-being. The project will also have direct educational impacts for groups underrepresented in STEM fields and generate anonymized datasets that other researchers can use.The team will develop experimental methods to reliably extract stress cues from commodity devices, using a suite of cognitive tasks that represent knowledge work and typical workplace stressors (e.g., time pressure, noise, distractions). Participants will perform the tasks and experience stressors while the team collects behavioral data from the commodity devices and ground truth stress measurements using physiological signals derived from thermal imaging. The team will evaluate how well features derived from the sensed behavioral data, using different sets of devices, can predict the ground truth stress data and how it varies based on specific stressors. The team will also develop a framework to deliver brief stress-reduction exercises that promote deep breathing, a proven effective and learnable stress reduction technique. The team will use iterative prototyping to develop novel, engaging mobile apps that use biofeedback, games, and music to support breathing exercises; these will be delivered by a multi-arm bandit-based recommendation system that considers the current context (predicted stress and stressors, time of day, particular computer activities) along with historical exercise adherence and results to suggest effective exercises. The stress sensing models and intervention framework will be validated through a series of lab and field studies with information workers at a software company, collecting stress data in situ with ecological momentary assessment techniques, validated survey instruments for stress and affect, and interviews.
