Modelo predictivo para identificar hogares beneficiarios de programas de transferencias monetarias: una comparación de técnicas de machine learning

Abstract

Cash transfer programs are a key tool for reducing poverty and improving the well-being of vulnerable households in developing countries. However, the accurate selection of beneficiaries remains a challenge. This study evaluates different machine learning techniques to predict participation in the Juntos program in Peru, using data from the 2023 National Household Survey (ENAHO). Models such as logistic regression, decision trees, support vector machine, gradient boosting machine, random forest, LightGBM, XGBoost, and CatBoost were compared. The results show that XGBoost achieves the best performance in beneficiary classification. These findings highlight the potential of machine learning techniques to enhance the allocation of resources in social programs. Their implementation would drive the modernization of public management, enabling data-driven economic resource management.