Understanding Stress In The Wild       

links

 photo

Understanding Stress In The Wild - overview


Stress has been classified as the "health epidemic of the 21st century” by the World Health Organisation. In a recent national survey by American Psychological Association, 75% of Americans reported experiencing at least one symptom of stress in the past month, such as anxiety and headaches. Numerous studies have suggested that stress can be associated with increased rate heart attack, hypertehtion, obesity, depression and many other physical and mental health problems. 

Given the possible role of stress in the causation and/or exacerbation of health problems, the ability to quantify and manage everyday stress becomes increasingly important. Our vision is to build a Cognitive IoT system that aggregates and analyzes available bio-signals to quantify and predict individual's stress level, and offers

• Timely awareness of elevated or prolonged stress;
• Personalized and actionable de-stress feedbacks;
• Analytics that uncover insights from historical data to inform various decision makings.
 
 
 

The challenges in understanding stress primarily lie in the fact that it is a highly personalized phenomenon that varies between individuals and different types of tasks. We have conducted a series of in-lab controlled studies where we collected an array of bio-signals (e.g., interbeat intervals, heart rate variability, respiration rate, etc.) from the participants while they were exposed to various types of physical and mental stressors. With the collected data, we created effective stress classifiers with multimodal learning, which essentially optimizes stress quantification by allowing a run-time integration of all relevant bio-signals and context information whenever they become available.

We have evaluated the stress model in real-world scinarios. The result based on the labeld data collected from corproate employees during multiple days suggest that the model can effectively provide real-time detection of daily stress (F1 score=76.9%) by considering the combination of Heart Rate Variability (HRV) parameters and context. In addition to everyday stress, we also tested the stress model in critical and demanding situations. We applied the model on interbeat interval data collected from firefighters during their training in a Rescue Maze suggests and demonstrated that the model is able to detect and differentiate mental and physical stress.