A Multi-level Analysis of Factors Associated with Student Performance: A Machine Learning Approach to the SAEB Microdata

School context matters more than individual circumstances: The study found that a school's average socioeconomic level was the strongest predictor of student performance—more influential than any single student or teacher characteristic. This suggests that systemic inequalities between schools are a primary driver of educational disparities in Brazil, highlighting the need for policies that address school-level disparities rather than focusing solely on individual student factors.

Machine learning can reveal hidden patterns in education data: By combining four different data sources (student backgrounds, teacher profiles, school indicators, and principal management), the researchers identified relationships that might not be obvious from traditional analysis. This multi-level approach demonstrates how AI can help policymakers understand complex educational systems more comprehensively.

Explainability is essential for responsible AI in policy: The researchers didn't stop at achieving high prediction accuracy (90.2%); they used Explainable AI (SHAP) to understand why the model made its predictions. This transparency is crucial for responsible AI use in education policy, ensuring that decisions affecting students are based on interpretable, understandable factors rather than "black box" algorithms.

Data-driven insights can guide equity-focused interventions: By identifying that school-level socioeconomic factors are dominant, this tool provides evidence to support targeted policies aimed at reducing educational inequality—such as directing resources to lower-performing schools or addressing systemic barriers within school ecosystems.