Stress Wearables - "Stress Detection Platform"

Project description

In our fast-paced world, stress poses a significant threat to overall well-being, impacting both mental and physical health with associated societal costs. To address this, wearable sensors have emerged as a important tool for accurate stress level measurement. These sensors enable individuals to monitor stress levels, benefiting personal well-being and aiding those who may struggle to communicate stress, such as individuals with dementia. Physiological signs like heart rate, sweating, and skin temperature, captured by wearables like smartwatches and chest straps, show a strong correlation with stress levels according to scientific research.

This project focuses on developing an application integrating wearable stress data into a versatile platform. Utilizing proven stress detection algorithms, the platform processes real-time data, presenting meaningful insights to end-users through tailored user interfaces. The goal is to empower individuals to manage stress effectively and alleviating the burdens of stress-related health issues.

Context

The main issue plaguing the project at the moment is that they have a decent amount of stress related data. However, they don’t have a way to display this information in a way that will indicate stress to a human. The stress related data is made up of a lot of different variables, on these variables an algorithm will need to be applied to, for data processing.

This processed data has to be put into a software system where an overview of each person wearing a wearable will be displayed. The overview should be able to display the stress factor of a given person, with the ability to change the sensitivity of the algorithm. All of this has to be monitored in case a stress related incident occurs.

Results

Our results have been divided into two domains, AI and software. Within the software domain an architecture framework design has been realized that supports the main requirements such as: "Real time streaming of wearable data", "Flexibility of integrating new wearables", "Stress detection through AI" etc. This design is based on the Event-Driven architecture pattern combined with Microservices. A prototype has been realized (TRL5) to study the architecture. This prototype is able to handle the real time streaming of data, which will be demoed.

In the Stress detection facet of our project, we explore the application of Anomaly detection on processed electrodermal activity (EDA) data using methods such as Z-score, Median, Quartile-based, Kernel Density, LOF (Local Outlier Factor), and MAD (Median Absolute Deviation). To substantiate our assertion regarding Stress detection within these metrics, we utilize the WESAD database, a well-established dataset in research papers. The validation results demonstrate that, notably, the LOF and MAD methods yield promising outcomes.

About the project group

Our group consists of two former Fontys Bachelor Software students, who are now studying for our Masters degree in Applied IT. We work closely together during the week having weekly scheduled meetings with our stakeholders to discuss our progress within the scope of the stress wearable project.