OBJECTIVE: Variables predicting obesity are not limited to individual-level risk factors. The purpose of this study is to assess multilevel predictors of obesity prevalence.
METHODS: US county-level datasets incorporating 34 variables were analyzed cross-sectionally using explainable artificial intelligence (XAI) analytical methods. A Light Gradient Boosting Machine Model was trained to predict obesity prevalence, after which model performance and feature importance were evaluated.
RESULTS: Optimal model performance included 29 features and explained 78% of the variance in county-level obesity prevalence. The dominant predictor of obesity prevalence was physical inactivity. Additional highly important variables include smoking, excessive drinking, political ideology, and regional culture.
CONCLUSIONS: This study used XAI methods to predict obesity, explaining 78% of the variance at the granular county level. Inclusion of both upstream and downstream factors in multisectoral and multidisciplinary approaches to predicting population-level obesity prevalence is warranted.