Stress can be detected during emotion-evoking smartphone use: a pilot study using machine learning.

Area of Science:

• Psychology
• Computer Science
• Biomedical Engineering

Background:

• Stress detection is crucial for timely intervention, but current methods have limitations.
• Contactless sensing using machine learning offers a potential alternative for stress estimation.
• Previous studies achieved high accuracy in classifying stress from facial expressions, but often used induced stress and classification approaches.

Purpose of the Study:

• To investigate if stress can be detected from facial expressions of six basic emotions and relaxation using a prediction approach.
• To explore associations between facial emotional expressions and stress levels.
• To compare the performance of Random Forest and XGBoost regression models for stress prediction.

Main Methods:

• Secondary analysis of video recordings of facial emotional expressions from 69 participants.
• Assessment of stress using the Perceived Stress Scale (PSS-10) and a one-item stress measure.
• Application of Random Forest (RF) and XGBoost regression models to predict stress scores from facial expression data.

Main Results:

• Facial emotional expressions were found to be promising indicators of stress scores.
• Model performance was optimal when data from all six emotional expressions were used for training.
• XGBoost demonstrated higher reliability and lower prediction error compared to RF for both training and test data.

Conclusions:

• Non-invasive video recordings of facial expressions can effectively complement standard objective and subjective stress markers.
• Machine learning models, particularly XGBoost, show potential for accurate stress prediction from facial cues.
• This research supports the use of contactless sensing for stress detection and intervention.