Stress Detection Using Wearable Physiological and Sociometric Sensors

Area of Science:

• Psychology
• Computer Science
• Biomedical Engineering

Background:

• Stress is a pervasive societal issue impacting individual well-being.
• Accurate, real-time stress detection is crucial for timely interventions.
• Current methods for stress assessment in social contexts have limitations.

Purpose of the Study:

• To develop and evaluate a machine learning approach for automatic stress detection in social settings.
• To investigate the efficacy of combining physiological and social sensor data for stress recognition.
• To assess the performance of various machine learning classifiers and identify key discriminative features.

Main Methods:

• Utilized a machine learning framework integrating data from two distinct sensor systems: one capturing physiological responses and the other social cues.
• Compared the performance of Support Vector Machine (SVM), AdaBoost, and k-Nearest Neighbor (k-NN) classifiers.
• Conducted experiments using a controlled Trier Social Stress Test (TSST) protocol.

Main Results:

• The combined sensor system significantly improved the accuracy of stress detection compared to individual modalities.
• Machine learning models demonstrated a high capability to discriminate between stressful and neutral conditions during the TSST.
• Analysis identified the most discriminative features contributing to accurate stress detection.

Conclusions:

• Combining physiological and social sensor data offers a robust approach for automated stress detection in social situations.
• The developed machine learning model shows promise for real-time stress monitoring applications.
• Further research into feature selection can optimize stress detection systems.