<![CDATA[The Nutritious Meal Program (MBG) aims to improve nutritional status and learning concentration among elementary school students; however, actual food consumption remains a critical challenge. This study develops a predictive model of daily MBG consumption based on students’ satisfaction and perceptions, and compares the performance of Random Forest and XGBoost algorithms. Data were collected from 132 students using a 10-item Likert-scale questionnaire and direct observation of plate waste. The target variable, “daily consumption,” was operationalized as an ordinal variable with three categories: 0 (rarely), 1 (sometimes), and 2 (often). Data were split using stratified sampling (80:20) and optimized through Grid Search with 5-fold cross-validation. The results indicate that XGBoost outperforms Random Forest, achieving lower MAE, MSE, and RMSE values, with statistically significant differences (p < 0.05). Feature importance analysis reveals that willingness to continue the program, perceived cleanliness, perceived taste, and grade level are the most influential factors affecting consumption. These findings highlight the critical role of students’ perceptions and attitudes in program effectiveness and demonstrate the potential of predictive modeling to support data-driven decision-making in school nutrition management. ]]>
Copyrights © 2026
