Bayesian Networks and Machine Learning Approaches Applied to Social Backwardness

Abstract This paper applies Bayesian and machine learning techniques to analyze Mexico’s Social Backwardness Index data from 2000 to 2020. This index aggregates key socioeconomic factors such as education, access to health services, essential housing services, housing quality and spaces, and household assets. We aim to identify the insights, such as conditional dependencies between these variables, and determine which factors most significantly contribute to social backwardness in Mexico. Through machine learning and non-parametric techniques (such as XGBoost, Neural Network Implementations, and Permutation Feature Importance), we identify which socioeconomic indicators most impact the degree of social backwardness. The Bayesian network is then employed to visualize the relationships between those socioeconomic indicators and the social backwardness index, providing information on the dependencies and linkages between features such as illiteracy, household appliances, and essential housing services. The analysis shows that critical indicators such as lack of household appliances, illiteracy, and inadequate housing services (e.g., lack of toilets and drainage) are highly predictive of social backwardness. Over the years, the importance of these variables shifts, but they remain consistently relevant in determining the level of social backwardness. Bayesian learning results suggest that policies targeting improvements in these primary household conditions could substantially reduce social backwardness across Mexico.