Ideb, as unidades da federação e o perfil das escolas públicas: uma análise exploratória de dados não supervisionada

Abstract

This dissertation seeks to use data science, through unsupervised exploratory analysis, to prove what the literature already knows and/or present new discoveries about the influence of infrastructure on Brazilian basic education. School infrastructure is not limited to the architectural issue of schools, but also to the educational and administrative environment, equipment, educational resources, practices, curricula and the teaching and learning process. Data collection was carried out on open data from the 2019 School Census (Basic Education) and the Basic Education Development Index (Ideb), for the years 2005-2019. The choice of the year 2019 was because it was the last year that schools presented results before the influence of the COVID-19 pandemic. After several data treatments and the choice of attending only the initial segment of fundamental education, two analysis methodologies were applied: Correlogram and Factor Analysis (FA). For clarity in the results, new attributes were created referring to the federative entities that allowed identifying which states and school profiles are better related to the growth and good results of the Ideb. For these correlations, the Sigma of the Gaussian Copula was chosen, which takes into account the categorical and continuous data and also generated a definite positive matrix. The Correlogram generated a square matrix that presented the attribute relationships in a Heatmap Dendrogram. Divided into 4 large groups, each one had specific characteristics and relationships with federal entities. The first group had a strong relationship with basic infrastructure; the second group, with IDEB and the most sophisticated infrastructures; the third group showed few relationships between the attributes; and the last group had strong negative correlations and contained greater precariousness in infrastructure. After verifying the compatibility of the database for the application of the FA, it was estimated that 10 factors would be suitable for this study. Four factors were associated with the attributes of the Ideb, the focus of this work. Three patterns were also observed in the attributes that listed good results in the Ideb with different infrastructures, policies and/or educational proposals: the first group, guided by São Paulo state, presented basic sanitation offered by the public service, quality internet for use in learning and institutions schoolchildren; the second group, headed by Minas Gerais state, indicates an association with flexibility in traditional teaching, with school cycles and non-serial classrooms; the third group was marked by complementary activities and specialized care, represented by the Ceará state. In contradiction to these parameters, schools with the EJA modality, mainly in the northeast, tend to have lower results in the Ideb. The other 6 factors added a lot of relevant information, including those related to the correlations and anti-correlations of the federative entities and specific attributes. As seen, data science has a lot to add to the field of education. Future works are expected to add even more data, such as longitudinal studies on the Ideb and to add other educational indices such as the Ioeb and the socioeconomic level of the population.