Explainable Artificial Intelligence (XAI) for bounded outcomes applied to the Municipal Human Development Index

Abstract

This study analyzes the Municipal Human Development Index (MHDI) of municipalities in the state of São Paulo using unit-distribution regression models and machine learning algorithms, including Random Forest and XGBoost. The methods are compared in terms of predictive performance and interpretability. Results indicate very similar predictive accuracy across approaches, suggesting that model performance is primarily driven by the information content of the predictors and that regression models for outcomes bounded in the unit interval remain competitive with more flexible machine learning methods. Per capita household income emerged as the most influential predictor of MHDI, followed by education and poverty-related variables. Model agnostic interpretability techniques revealed nonlinear effects and provided both local and global insights into predictor contributions. The findings offer a robust and interpretable framework for analyzing municipal development and provide evidence that may support regional monitoring and public policy planning.