Evaluation and Comparison of Three Mixed-Effect Models for Household Food Insecurity Classification in West Java

Abstract

Food insecurity is a complex and multidimensional issue that requires accurate predictions or classifications to design effective interventions. The availability of large and complex datasets has enabled the use of machine learning approaches. Moreover, because food insecurity data are often hierarchical/clustered, mixed effects modelling is well suited for capturing intergroup variation. This study compared three models: the generalized linear mixed model (GLMM), a parametric model that accounts for random effects but is limited to linear relationships; the generalized mixed effects tree (GMET), a flexible decision tree framework with random effects; and the generalized mixed effects random forest (GMERF), an ensemble of trees with random effects. The analysis used data from 25,890 households in West Java, Indonesia in 2021. Model evaluation showed that GMERF provided the best prediction results compared to the other models. This study concludes that integrating random forests with mixed-effects modelling offers a robust and effective approach for predicting household food insecurity. The primary predictors of food insecurity identified by the random forest in GMERF model were the age of the household head (years) and house size (m2).