AI model reveals physical inactivity as the top driver of obesity across US counties

A new study published in Obesity used a machine learning model to analyze what's really driving obesity rates across the US — and the answer goes well beyond diet. Researchers examined 34 county-level variables spanning culture, politics, social environments, behaviors, and health characteristics across 2,284 US counties. The best-performing model (a Light Gradient Boosting Machine) explained 78% of the variation in county-level obesity prevalence, a remarkably strong result for a population-level analysis.

Physical inactivity came out on top as the strongest predictor, with higher inactivity rates linked to a jump in predicted obesity prevalence from ~34.5% to ~41.5%. Smoking, excessive drinking, cultural grouping, insufficient sleep, and political ideology rounded out the top six predictors. A simulated 10% reduction in physical inactivity lowered predicted obesity in some counties — but regional hotspots actually showed an increase, highlighting the complexity of the issue.

By the Numbers

• 78% of county-level obesity variance explained by the model
• 29 out of 34 variables used in the final model
• Obesity prevalence range tied to inactivity: ~34.5% → ~41.5%
• 10% inactivity reduction → up to ~2% drop in predicted obesity in some counties

Why it matters: This AI-driven approach shows that tackling obesity requires looking beyond individual choices to broader social and environmental forces — and that boosting physical activity at the population level could be one of the most impactful levers policymakers have.