Session 7 : Using Machine Learning to Predict Attrition in a Federal Nutrition Education Program
Presentation Type
Oral
Student
No
Track
Other
Abstract
Attrition poses a significant challenge to the effectiveness of federal nutrition education programs, hindering their ability to achieve widespread impact. This study employs machine learning techniques to develop a predictive model for identifying participants at high risk of dropping out of the Expanded Food and Nutrition Education Program (EFNEP). Analysis is conducted using standardized EFNEP program data (pre-program 24-hour dietary recalls, food and physical activity questionnaires, and demographic information) on over 1.25 million adult participants from 2013 to 2022. Three machine learning algorithms (logistic regression, XGBoost, and random forest) were evaluated, with the XGBoost model achieving the highest predictive accuracy. Key predictors of attrition included Cooperative Extension region, funding tier, land-grant university type (1860 vs. 1890), enrollment year, household income, age, race, residence, number of children, number of foods consumed in 24-hour dietary recall, and physical activity level. These findings provide valuable insight to EFNEP administrators, enabling them to proactively identify at-risk program participants and implement targeted interventions to improve retention and program impact.
Start Date
2-7-2025 2:30 PM
End Date
2-7-2025 3:30 PM
Session 7 : Using Machine Learning to Predict Attrition in a Federal Nutrition Education Program
Jacks' Place (Room 050)
Attrition poses a significant challenge to the effectiveness of federal nutrition education programs, hindering their ability to achieve widespread impact. This study employs machine learning techniques to develop a predictive model for identifying participants at high risk of dropping out of the Expanded Food and Nutrition Education Program (EFNEP). Analysis is conducted using standardized EFNEP program data (pre-program 24-hour dietary recalls, food and physical activity questionnaires, and demographic information) on over 1.25 million adult participants from 2013 to 2022. Three machine learning algorithms (logistic regression, XGBoost, and random forest) were evaluated, with the XGBoost model achieving the highest predictive accuracy. Key predictors of attrition included Cooperative Extension region, funding tier, land-grant university type (1860 vs. 1890), enrollment year, household income, age, race, residence, number of children, number of foods consumed in 24-hour dietary recall, and physical activity level. These findings provide valuable insight to EFNEP administrators, enabling them to proactively identify at-risk program participants and implement targeted interventions to improve retention and program impact.