An interpretable machine learning model of cross-sectional U.S. county-level obesity prevalence using explainable...

Area of Science:

• Public Health
• Data Science
• Epidemiology

Background:

• Obesity prevalence exhibits significant geographic disparities across US counties.
• Machine learning models accurately predict these variations but often lack interpretability.
• Understanding the drivers of geographic obesity variation is crucial for targeted interventions.

Purpose of the Study:

• To extract actionable knowledge from machine learning models regarding county-level obesity prevalence variation.
• To enhance the interpretability of predictive models for public health applications.
• To identify key factors influencing obesity across diverse geographic regions in the US.

Main Methods:

• Applied explainable artificial intelligence (XAI) methods to machine learning models predicting obesity.
• Utilized cross-sectional obesity prevalence data from 3,142 US counties.
• Incorporated county-level features across health outcomes, behaviors, clinical care, socioeconomics, environment, demographics, and housing.

Main Results:

• Machine learning models explained 79% of the variance in county-level obesity prevalence.
• Physical inactivity, diabetes prevalence, and smoking prevalence were identified as the most significant predictors.
• Feature importance and other XAI techniques elucidated the contributions of various factors.

Conclusions:

• Interpretable machine learning models offer substantial insights into the geographic variation of obesity prevalence.
• Understanding the interplay of health behaviors and outcomes is key to addressing obesity disparities.
• XAI facilitates the translation of complex models into practical public health knowledge.