Related Experiment Video
Updated: Jul 9, 2026
05:59
Visualization of Intensity Levels to Reduce the Gap Between Self-Reported and Directly Measured Physical Activity
Published on: March 7, 2019
Understanding Drivers of Physical Activity Through Explainable Machine Learning: The Role of Disability and Social
Aditya Chakraborty1, Emily J Nicklett2
1Department of Epidemiology, Biostatistics, and Environmental Health, Joint School of Public Health, Old Dominion University, Norfolk, VA, USA.
Journal of Physical Activity & Health
|July 7, 2026
Summary
Machine learning models accurately predict exercise participation in adults with disabilities. Mobility, income, and education are key factors influencing activity levels, especially within different disability groups.
Area of Science:
- Public Health
- Biostatistics
- Artificial Intelligence
Background:
- Physical activity in adults with disabilities is influenced by functional limitations, health, and socioeconomic factors, but their predictive importance is unclear.
- Understanding these predictors is crucial for developing targeted interventions to promote exercise participation.
Purpose of the Study:
- To compare machine learning models for predicting exercise participation among adults with disabilities.
- To identify the most influential predictors of exercise participation using explainable artificial intelligence.
Main Methods:
- Utilized the Centers for Disease Control and Prevention Behavioral Risk Factor Surveillance System data, including 5 disability indicators and various demographic, socioeconomic, and health factors.
- Compared five machine learning models (logistic regression, LASSO, SVM, random forest, XGBoost) and evaluated performance using AUC, accuracy, F1 score, sensitivity, and specificity.
- Employed SHAP (SHapley Additive exPlanations) plots for model explainability to identify key predictors.
Main Results:
- XGBoost (AUC=0.83) and random forest (AUC=0.80) showed the strongest predictive performance for exercise participation.
- A significant socioeconomic gradient was observed, with lower income and education correlating with reduced exercise participation across disability types.
- SHAP analyses identified mobility status, education, income, physical health, and age as the top predictors of exercise participation.
Conclusions:
- Explainable machine learning effectively identifies individuals at higher risk of physical inactivity.
- Disability type, socioeconomic status, and individual characteristics significantly interact to influence exercise participation.
