Machine Learning Frameworks for Wearable-Based Stress Modeling in Naturalistic Settings: Scoping Review

Area of Science:

• Wearable technology
• Machine learning
• Mental health assessment

Background:

• Modern life's pervasive stress impacts mental and physical health.
• Wearable devices excel at physical fitness tracking, but mental stress detection is nascent.
• Objective assessment of stress using technology remains a significant challenge.

Purpose of the Study:

• To review recent studies on wearable-based stress detection in naturalistic settings.
• To characterize machine learning frameworks used in these studies, inspired by the model card approach.
• To identify trends, challenges, and propose standardized reporting for stress detection research.

Main Methods:

• Systematic scoping review following PRISMA-ScR guidelines.
• Searched major scientific databases (PubMed, IEEE, ACM, etc.) for relevant articles.
• Included studies using wearable devices and machine learning for stress detection in healthy adults in naturalistic settings.

Main Results:

• Analyzed 34 eligible studies published between 2017 and 2024.
• Examined machine learning decisions: problem formulation, ground truth, algorithms.
• Proposed a model card framework for reporting wearable-based stress detection models.

Conclusions:

• Highlights trends in machine learning for wearable-based stress detection.
• Emphasizes the need for standardized reporting of datasets and ML decisions.
• Underscores challenges in real-world data collection and promotes collaborative research advancement.