ANALYSIS OF DECISION TREE-BASED PREDICTIVE MODELS FOR STUNTING PREVALENCE IN JAVA USING SOCIO-ENVIRONMENTAL PARAMETERS

Abstract

Stunting prevalence indicates the percentage of children under five whose height falls below the WHO standard. As the most densely populated region in Indonesia, Java had a critical role in policy formulation related to stunting reduction. This study aimed to develop a predictive model of stunting prevalence in Java influenced by social and environmental factors using tree-based machine learning approaches, Classification and Regression Tree (CART), Random Forest, and Extreme Gradient Boosting (XGBoost). Simulations were conducted using these three models, and model performance was evaluated using Root Mean Square Error (RMSE). Feature selection based on feature importance values was applied with proportions of 25%, 50%, 75%, and 100% of the features. The results indicated that XGBoost achieved the best performance with a mean RMSE of 5.3820 using 50% of the features and demonstrated the highest prediction stability. Across the best-performing configurations of each model, four features were consistently selected: Posyandu Activities, Toddler Mothers Class, Caregiving Class, and Utilization of Family and Village Yard Land. These findings indicated that strengthening interventions in caregiving practices, community-based health education services, and environmental resource utilization had the potential to become priority programs in efforts to reduce stunting prevalence in Java.