Evaluation of machine learning models based on household food insecurity data in Indonesia

Abstract

Household food insecurity is a critical issue, and accurate prediction models are essential for identifying at-risk households and guiding policy decisions. This study compares the effectiveness and stability of two machine learning models, Random Forest (RF) and Generalized Random Forest (GRF). The predicting household food insecurity using Food Insecurity Experience Scale data in West Java, Indonesia. The evaluation shows that the GRF model performs best and is more consistent. Key predictors in this analysis include household size and type of house flooring, along with other variables such as household savings, type of house walls, sanitation facilities, cash transfer program status, land ownership, and food assistance recipient status.