Accurate Chronic Stress Estimation with Personalized Models Based on Correlation Maximization.

Area of Science:

• Physiological computing
• Psychological assessment
• Machine learning for healthcare

Background:

• Chronic stress significantly impacts health and well-being.
• Accurate estimation of chronic stress is challenging due to individual variability.
• Existing methods for stress estimation lack personalization and sufficient accuracy.

Purpose of the Study:

• To develop an accurate chronic stress estimation system using personalized models.
• To identify effective physiological features for stress estimation through correlation maximization.
• To classify individuals into groups where feature-ground truth relationships differ.

Main Methods:

• Personalized models trained on physiological features correlated with ground truth (Perceived Stress Scale).
• Classification of users into distinct groups based on personality questionnaires and conventional methods.
• Evaluation using 1-month wearable sensor data from 168 subjects.

Main Results:

• The proposed classification system achieved 69.1% accuracy in estimating chronic stress (increase/decrease in PSS).
• This surpasses conventional methods: 59.3% (no classification) and 56.8% (k-means clustering).
• Identified specific physiological features effective for personalized stress estimation.

Conclusions:

• Personalized modeling significantly enhances chronic stress estimation accuracy.
• Individual classification based on personality traits improves model performance.
• The developed system offers a more accurate approach to monitoring and managing chronic stress.