Research on Income Assessment of Poor Families in Latin America Based on Machine Learning Algorithms

This study tackles the critical challenge of accurately evaluating incomes in economically vulnerable households by constructing a machine learning framework utilizing Costa Rican household data. The proposed methodology emphasizes systematic feature engineering to identify socioeconomic attributes with strong statistical correlations to income levels, including housing conditions, educational attainment, and employment patterns. By prioritizing these discriminative features through rigorous correlation analysis and dimensionality reduction techniques, the framework substantially optimizes input data quality for predictive modeling. Comparative experiments demonstrate that this feature selection strategy consistently enhances the performance of diverse baseline algorithms, including decision trees, random forests, and gradient boosting machines, achieving notable improvements in prediction accuracy and generalization capability. The findings highlight the importance of domain-informed feature optimization in developing robust socioeconomic assessment tools, offering valuable insights for policymakers designing targeted poverty alleviation programs. This data-driven approach provides a scalable solution for income estimation challenges in developing regions, particularly where traditional survey methods face implementation barriers.