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

A machine learning analysis of Brazil’s national student assessment data has found that a school’s average socioeconomic level is the strongest predictor of academic performance, outweighing individual student or teacher characteristics, according to a study posted on arXiv [1][2]. The study, authored by Rodrigo Tertulino and submitted in October 2025 before being revised in June 2026, used microdata from the System of Assessment of Basic Education, known by its Portuguese acronym SAEB [1][2]. Researchers integrated four data sources: student socioeconomic characteristics, teacher professional profiles, school indicators, and principal management profiles [2]. The work focused on 9th-grade and high school students [2]. A Random Forest model outperformed three other ensemble algorithms, reaching 90.2% accuracy and an Area Under the Curve of 96.7% [2]. To interpret the model’s decisions, the team applied SHAP, an explainable AI technique [2]. The analysis showed that the school’s average socioeconomic level was the most dominant predictor, leading the authors to conclude that systemic factors carry more weight than individual characteristics in isolation [2]. The paper describes academic performance as “a systemic phenomenon deeply tied to the school’s ecosystem” and frames the model as a data-driven tool for policies aimed at reducing disparities between schools [2]. The research appears on arXiv, an open-access repository that hosts electronic preprints across disciplines including computer science and statistics [6]. As of November 2024, arXiv was receiving about 24,000 new submissions per month and had surpassed two million total articles by the end of 2021 [6]. Preprints on arXiv are moderated but not peer-reviewed, a distinction the repository maintains while providing rapid dissemination [6]. The platform also supports experimental community tools through its arXivLabs framework, which allows collaborators to build features such as citation explorers and code finders that appear on article pages [4][5]. These integrations operate under guidelines that require partners to uphold openness, community engagement, and user data privacy [4].