Leveraging Causal Reasoning in Educational Data Mining: An Analysis of Brazilian Secondary Education

Abstract

This study presents an approach to investigating the main interventions related to gains on performance using a combination of educational data mining (EDM) techniques and traditional theory-driven models. The goal is to overcome the limitation of previous EDM studies that lack of causal reasoning, which is a critical concern for educational specialists. We use large-scale assessment data from Brazil and map the main sources of unobserved confounders using causal graphs. We then use a two-way logistic regression fixed effects to account for these confounding factors. The model is evaluated for its predictive ability and further investigated through classification rules and decision trees, resulting in the proposition of new insights into the data. The findings of the study underline the importance of socio-economic factors and showcase the significant impact of faculty education policies as well as the vital role of Brazilian states in these policies.

Keywords

• educational data mining
• educational assessment
• causality
• ENEM