<![CDATA[Food anxiety represents an early psychological indicator of household food insecurity and is influenced by economic vulnerability, household characteristics, and unstable access to food. West Java, as Indonesia’s most populous province, faces substantial socio-economic disparities that heighten the risk of food insecurity. Using SUSENAS 2024 data, this study aims to classify household food anxiety and evaluate the predictive performance of three boosting algorithms XGBoost, LightGBM, and CatBoost. The dataset exhibits a strong class imbalance, with only 19.1% of households categorized as food anxious, prompting the application of SMOTE and Winsorization during preprocessing. SMOTE considerably improved model performance, particularly in balanced accuracy. For XGBoost, balanced accuracy increased sharply from 0.5199 to 0.8738, while LightGBM experienced a similar improvement from 0.5261 to 0.8736. Winsorization produced only marginal additional effects. Across all scenarios, XGBoost demonstrated the highest overall performance, followed closely by LightGBM, whereas CatBoost showed limited ability to detect minority-class households. These findings underscore the effectiveness of boosting algorithms especially XGBoost enhanced by SMOTE in identifying food-anxious households and supporting data-driven, targeted food security interventions in West Java.]]>
Copyrights © 2026
