<![CDATA[Accurately identifying food-insecure households is challenging because the condition is inherently multidimensional. This study compares three machine-learning approaches — Random Forest (RF), Double Random Forest (DRF), and RF combined with EasyEnsemble class balancing — for predicting household food insecurity in Langsa City. Data from the 2024 SUSENAS survey cover 2,057 households with 13 predictor variables. A berat_count ≥ 3 threshold on the FIES indicators defines the target variable, yielding 755 food-insecure (36.7%) and 1,302 food-secure (63.3%) households. Fifty repetitions with a 70:30 train-test split yield stable performance estimates. RF + EasyEnsemble achieves the best results with a mean AUC of 0.8398 and sensitivity of 79.84%, far surpassing DRF at 2.62%. ANOVA (F = 191.899; p 0.001) and Tukey HSD tests confirm statistically significant differences. Feature importance reveals social-assistance participation (54.78%) and physical housing conditions (28.89%) as the dominant predictors.]]>
Copyrights © 2026
