< Back to previous page


Stress reduction at work using wireless sensors.

In the 21st century, stress and mental health have become major concerns worldwide. Yet, a continuous, quantitative measurement technique, allowing just-in-time interventions to reduce stress, is lacking. Therefore, research has focused on exploiting the sympathetic nervous system’s (SNS) fight-or-flight response, by investigating physiological signals for monitoring stress. Research has focused on developing machine learning models for stress detection, based on physiological signals such as heart rate (HR), skin conductance (SC), skin temperature (ST) and respiration. These have shown to be reliable indicators of stress in well-controlled laboratory conditions, but large-scale ambulatory validation is missing.

The goal of this research was to identify physiological sensing priorities and machine learning techniques for physiological stress detection and next, to deploy these on a large population in real-life, ambulatory conditions.

First, we identified the most suitable markers for physiological stress detection. We concluded that, on average, SC and HR related features are more important than ST and respiration related features. However, on a personal level, physiological sensing priorities differ across subjects, favoring a multi-sensor approach. The selection of the most optimal machine learning technique depends on the context of the application.

Second, we were able to differentiate between healthy subjects and patients based on their physiological stress response with an accuracy of 78%, which is a promising result towards disease prevention and interception.

Third, we presented the SWEET study: world’s largest ambulatory stress detection study, including 1,002 subjects who were continuously monitored during 5 days. We presented a methodology to use subject-specific information, based on the physiological response to a specific stress task in an ambulant environment, towards a personalized calibration for ambulant stress detection models.

The results of this dissertation provide a first step towards personalized stress detection, and more generally towards precision medicine and personalized healthcare.

Date:31 Mar 2014 →  31 Dec 2018
Keywords:Psychological stress, psychophysiology, workplace, machine learning
Disciplines:Nanotechnology, Design theories and methods
Project type:PhD project