Predicting Attrition in a Public Nutrition Education Program: A Machine Learning Approach

OBJECTIVE: To develop a machine learning model to identify adult Expanded Food and Nutrition Education Program (EFNEP) participants at high risk of attrition using preprogram data. DESIGN: Secondary data analysis. PARTICIPANTS: A total of 339,335 adults participating in EFNEP nationwide from 2013-2022. MAIN OUTCOME MEASURES: Adult EFNEP attrition, defined as a participant dropping out of EFNEP before the end of the program. ANALYSIS: Three machine learning models (logistic regression, random forest, and eXtreme Gradient Boosting) were developed and evaluated to predict attrition based on participants' preprogram food behaviors, dietary intake, demographics, and program characteristics. RESULTS: The eXtreme Gradient Boosting model demonstrated the best predictive performance, achieving the highest F1 score of 0.68 (a measure of the balance between correctly identifying true dropouts and avoiding false positives), and a recall of 70% (correctly identifying dropouts). Key predictors of attrition included Cooperative Extension region, EFNEP funding tier, enrollment year, household income, age, race, residence, number of children, and preprogram dietary intake and physical activity. CONCLUSIONS AND IMPLICATIONS: Application of machine learning models in nutrition education program settings can uncover previously unknown patterns contributing to attrition, informing targeted interventions and resource allocation to improve retention and maximize program impact.