Stress Detection in Video Feed: Utilizing Facial Action Units as Indicators in Various Machine Learning Algorithms

Stress has always been present in our daily lives. This has and always been essential in helping us grow. Yet, once stress becomes too much to handle and exhausts an individual to a point where there is no space for recovery, this stress may develop into the chronic stage. This harmful state of stre...

Full description

Saved in:
Bibliographic Details
Main Authors: Llanes, Rizzah Grace, Reyes, Rosula SJ
Format: text
Published: Archīum Ateneo 2022
Subjects:
Online Access:https://archium.ateneo.edu/ecce-faculty-pubs/135
https://doi.org/10.1109/ECBIOS54627.2022.9945040
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Ateneo De Manila University
Description
Summary:Stress has always been present in our daily lives. This has and always been essential in helping us grow. Yet, once stress becomes too much to handle and exhausts an individual to a point where there is no space for recovery, this stress may develop into the chronic stage. This harmful state of stress may lead to vulnerabilities in physical and mental health. It can also affect one's quality of life and can hinder productivity in their everyday tasks and responsibilities in school or at work. Multiple studies have been done to detect stress by gathering physiological signals but these invasive measures may be affected by multiple factors. This research aimed to detect stress using a non-invasive measure of detecting facial actions units in the video setting. This was done by taking a recorded footage of the participant while answering an arithmetic test in a time-limited and competitive environment. Stress levels were validated through the participant's own evaluation of how stressful the test was. Extraction of the facial action units was done by utilizing the OpenFace 2.0 interface. The following machine learning algorithms were used to classify the stress levels: multiple linear regression, support vector machine. Classifier performance was measured through accuracy and F1 score and found that random forest model performed the best in the classification using overall data and person specific data.