Forecasting stress transitions using ecological momentary assessment data and machine learning.

Area of Science:

• Digital Health
• Computational Psychiatry
• Machine Learning in Mental Health

Background:

• Stress negatively impacts sleep, cognition, and mental health.
• Ecological Momentary Assessments (EMAs) capture real-time experiences in natural settings.
• Just-in-time adaptive interventions (JITAIs) leverage EMA data for timely support.

Purpose of the Study:

• To forecast stress levels using machine learning and EMA data.
• To enable proactive, personalized stress management interventions.
• To evaluate model performance across diverse European populations.

Main Methods:

• Formulated stress forecasting as a binary classification problem (normal vs. elevated stress).
• Utilized EMA data from vulnerable groups (youth, older adults, migrants, low SES).
• Trained and evaluated machine learning models for stress transition prediction.

Main Results:

• Machine learning models achieved 0.70 ROC-AUC for stress forecasting.
• Predicting transitions to elevated stress was more challenging than stable normal stress.
• Models trained on combined data performed comparably to population-specific models.

Conclusions:

• Machine learning effectively forecasts stress transitions, supporting proactive mental health interventions.
• Forecasting models show generalizability across diverse European populations.
• Personalized stress prediction using EMA data is feasible and beneficial.