IDEB, THE FEDERATIVE UNITS AND THE SCHOOL CENSUS: AN UNSUPERVISED ANALYSIS

• Autor(a) principal: Alves, Igor
• Data de Publicação: 2024
• Outros Autores: Jesus, Rodrigo, Romanelli, João, Silveira, Carlos
• Tipo de documento: preprint
• Idioma: por
• Título da fonte: SciELO Preprints
• Texto Completo: https://preprints.scielo.org/index.php/scielo/preprint/view/6097
• topic: Indicadores educacionais; ensino fundamental; análise exploratória de dados; modelos não supervisionados; Educational indicators; elementary education; exploratory data analysis; unsupervised models; Indicadores educativos; educación primaria; análisis exploratorio de datos; modelos no supervisados

This multidisciplinary work proposes new perspectives and reflections on the data from the multifaceted reality of Brazilian education. It was based on a meticulous methodology of unsupervised exploratory analysis, without any expectation of results, in order to allow the emergence of new insights spontaneously from the data. Special attention was given to data visualization in a multidimensional context, as a way to sensorially highlight the most immanent patterns, without compromising statistical rigor. Therefore, the main objective of this research was to create an innovative intertwining of techniques that, in the context of educational data, would offer more opportunities for questions and inquiries than answers. The database consisted of the Basic Education Development Index (Ideb) and the School Census, within a scope limited to elementary education in 58,920 public schools across Brazil. At the heart of the methodology was a Gaussian Copula paired with a recursive Factor Analysis, after careful treatment of missing values, outliers, linear dependencies, and feature selection. As a result, a plurality of patterns was obtained, both known from the literature and unprecedented. Like Ideb having weight in more than one dimension, implying a heterogeneity of associations involving this index, the infrastructure attributes, and the federal units. In conclusion, the methodology has shown to be reliable and robust in the challenge of composing new analytical perspectives on educational data, enabling the expansion of the scope in future studies.