Use of machine learning for predicting stress episodes based on wearable sensor data: A systematic review

Area of Science:

• Biomedical Engineering
• Computer Science
• Psychology

Background:

• Stress monitoring is crucial for mental and physical well-being.
• Traditional stress assessment methods are often subjective and time-consuming.
• Wearable technology offers continuous, objective physiological data collection.

Purpose of the Study:

• To systematically review the integration of wearable sensor data and machine learning (ML) for stress prediction.
• To identify prevalent sensors, key physiological features, and ML model effectiveness.
• To explore the potential for real-world stress monitoring and prediction.

Main Methods:

• Systematic literature review following PRISMA methodology (January 2010 - June 2025).
• Data sources: IEEE Xplore, Elsevier, Springer, MDPI, PubMed Central, ACM.
• Inclusion criteria: Wearable devices, ML stress prediction, physiological data (HRV, skin conductance, sleep).

Main Results:

• Diverse sensors (EDA, PPG, accelerometers) and features (HRV, skin conductance, respiratory patterns) are utilized.
• ML models (RF, SVM, DNN) show high predictive accuracy (up to 99%).
• Challenges include small sample sizes, data variability, and lack of standardized protocols.

Conclusions:

• Wearable sensors and ML offer scalable, real-time stress monitoring for proactive healthcare.
• Advancement requires addressing limitations via interdisciplinary collaboration and improving technology accessibility.
• This field has transformative implications for personalized health, mental health support, and healthcare efficiency.