Loading...
Using Explainable Artificial Intelligence to Understand Cross-sectional Obesity Prevalence in Rural and Urban U.S. Counties
Wood, Rebecca
Wood, Rebecca
Citations
Altmetric:
Abstract
Rural and urban environments can present distinct exposures that increase the risk for obesity. Machine learning approaches offer accurate models for geographic differences and obesity prevalence, but the models are often uninterpretable black boxes. The study uses SHapley Additive exPlanations, an explainable artificial intelligence approach, to perform a comparative analysis of two different machine learning models to better understand the risks for obesity in these unique environments. Utilizing data from the 2023 County Health Rankings, an aggregation of health-related metrics for 3,133 U.S. counties, we trained both models to predict obesity prevalence, using fivefold cross-validation. Model performance was based on the main absolute error of predictions, and distinct predictors for rural and urban areas were based on SHapley Additive exPlanations for future contributions in those counties. XGBoost outperformed Random Forest with mean absolute errors of 1.55 and 1.68, respectively. Notably, both models predicted obesity more accurately in rural than in urban areas (errors of 1.43 and 1.82, respectively). SHapley Additive exPlanations showed that the key factors influencing obesity prevalence across both environments included physical inactivity, smoking prevalence, and solo driving. Low air pollution was associated with lower obesity in rural areas, whereas a greater housing cost burden in urban areas predicted lower obesity prevalence. This study highlights the unique contributors to obesity and rural versus urban environments and reveals the general role of physical activity and obesity prevalence. The explainable artificial intelligence approach to machine learning can serve as a foundation for developing obesity treatments that are tailored to insights for models of regional and patient-level characteristics.
Description
These are the slides from a presentation given at the Obesity Week held in San Antonio, TX on 11/04/2024.
Date
2024-11-04
Journal Title
Journal ISSN
Volume Title
Publisher
University of Kansas
Research Projects
Organizational Units
Journal Issue
Keywords
Rural and Urban U.S. Counties, Obesity Prevalence, XGBoost