Food insecurity is a major public health concern in the United States given its association to diet quality and consequently health outcomes. Routine county-level estimates of food insecurity have played a key role in informing intervention programs, policies, and monitoring progress. Although the definition of food insecurity encompasses access to nutritious food, estimates of food insecurity are commonly based on metrics that focus on access to adequate food and not necessarily nutritious food. Though people experiencing food insecurity are also likely to be nutrition insecure, there is growing interest in monitoring broader nutrition security, to identify context-specific contributing factors and better inform suitable intervention programs. This exploratory research uses data from national-level surveys and machine-learning to develop predictive models for food and nutrition insecurity to inform predictive models for county-level estimates.

Data from the National Health Interview Survey (NHIS) and the National Health and Nutrition Examination Survey (NHANES) was preprocessed and analyzed through the Science Collaborative for Health and Artificial Intelligence Reduction of Errors (SCHARE) on the Terra cloud platform. eXtreme Gradient Boosting (XGBoost) models were trained for each dataset to classify individuals as food and nutrition secure or insecure based on identified socioeconomic and health predictors. Hyperparameters were tuned to maximize F1 scores, ensuring high classification performance for these unbalanced datasets. Shapely Additive exPlanations (SHAP) were utilized to identify dominant predictors for each model and multicollinearity was assessed with Spearman’s ρ. Expected results include identifying important predictors of nutrition security and a comparative analysis with features influencing food security.