A Micro-Macro Machine Learning Framework for Predicting Childhood Obesity Risk Using NHANES and Environmental Determinants

TL;DR

This paper tackles the challenge of predicting childhood obesity by bridging micro-level NHANES data with macro-level environmental determinants from USDA and EPA. It introduces a micro-macro ML framework that trains four classifiers on NHANES, builds a state-level EnvScore to summarize structural burdens, and performs cross-scale alignment using visualization and clustering. The strongest micro-model (XGBoost) achieves an AUC of $0.714$, while EnvScore reveals pronounced geographic heterogeneity that co-occurs with higher predicted obesity risk, suggesting environment-driven disparities. The approach enables scalable public health informatics, offering a practical pathway for targeted interventions and informing multi-sector policy, with potential extensions to causal inference and real-time analytics.

Abstract

Childhood obesity remains a major public health challenge in the United States, strongly influenced by a combination of individual-level, household-level, and environmental-level risk factors. Traditional epidemiological studies typically analyze these levels independently, limiting insights into how structural environmental conditions interact with individual-level characteristics to influence health outcomes. In this study, we introduce a micro-macro machine learning framework that integrates (1) individual-level anthropometric and socioeconomic data from NHANES and (2) macro-level structural environment features, including food access, air quality, and socioeconomic vulnerability extracted from USDA and EPA datasets. Four machine learning models Logistic Regression, Random Forest, XGBoost, and LightGBM were trained to predict obesity using NHANES microdata. XGBoost achieved the strongest performance. A composite environmental vulnerability index (EnvScore) was constructed using normalized indicators from USDA and EPA at the state level. Multi-level comparison revealed strong geographic similarity between states with high environmental burden and the nationally predicted micro-level obesity risk distribution. This demonstrates the feasibility of integrating multi-scale datasets to identify environment-driven disparities in obesity risk. This work contributes a scalable, data-driven, multi-level modeling pipeline suitable for public health informatics, demonstrating strong potential for expansion into causal modeling, intervention planning, and real-time analytics.

Figures (10)

• Figure 1: SHAP interpretability summary plot illustrating feature contributions to micro-level childhood obesity predictions. Age, household income ratio, and race/ethnicity show the strongest marginal effects, consistent with pediatric epidemiology.
• Figure 2: Correlation matrix of U.S. state-level socioeconomic, food access, and air-quality indicators derived from USDA Food Access Research Atlas and EPA AQS summaries. Variables include poverty rate, median family income, food desert prevalence, and multiple air-quality metrics.
• Figure 3: Relationship between limited food access (LILAT%) and annual ozone pollution days. States with constrained food environments also tend to experience greater air pollution burden, demonstrating compound environmental stressors relevant to childhood obesity.
• Figure 4: Unsupervised clustering of states using poverty, food access, and air-quality profiles. Cluster 2 (green) represents high-vulnerability states, often overlapping with regions where childhood obesity prevalence has been reported as higher in prior epidemiological work.
• Figure 5: Mean predicted childhood obesity risk by U.S. state using the NHANES-trained machine learning model.