<![CDATA[Food insecurity remains a serious household-level issue, particularly in densely populated regions such as West Java, highlighting the need for analytical approaches capable of accurately identifying vulnerable groups. Machine learning algorithms offer the potential to improve the accuracy and precision of food insecurity classification based on survey data. This study aims to compare the predictive performance and variable importance identification of four machine learning algorithms—Random Forest, Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN)—in predicting household food insecurity status. The analysis employs SUSENAS 2023 data covering 26,012 households with 14 predictor variables, and food insecurity is classified using the Food Insecurity Experience Scale (FIES). Class imbalance is addressed using the Synthetic Minority Over-sampling Technique (SMOTE) within a 10-fold cross-validation framework. The results show that XGBoost achieves the highest accuracy of 71%, while Random Forest provides the best balanced accuracy under the SMOTE scenario. Moreover, all algorithms consistently identify the Wealth Index as the most influential predictor based on their respective Variable Importance measures, followed by variables related to water access and food assistance. Accordingly, XGBoost is recommended in terms of accuracy, whereas Random Forest demonstrates superior balanced accuracy and prediction stability.]]>
Copyrights © 2026
