Análise de matrículas escolares em Maceió: um estudo de séries temporais com LSTM

Abstract

The prediction of school enrollment trends is essential for educational planning, enabling the efficient allocation of resources and the formulation of data-driven public policies. This study analyzes the factors influencing enrollment rates in preschool, elementary, and high school education in the public network of Maceió, using Long Short-Term Memory (LSTM) neural networks for time series analysis. The research was conducted with data from the INEP School Census, covering a ten-year period. In addition to predictive analysis, Explainable Artificial Intelligence techniques such as SHAP (SHapley Additive Explanations) and Granger Causality were applied to interpret the most relevant factors affecting enrollment trends. The results indicate that, in preschool education, school infrastructure, including adapted restrooms, multimedia equipment, and playgrounds, directly impacts parents’ decisions regarding enrollment. In elementary school, maintaining an adequate teaching staff and providing well-structured libraries were identified as crucial factors for student retention. In high school, the availability of science laboratories, sports infrastructure, and access to technology proved to be essential for student retention. Furthermore, the analysis revealed interdependencies among variables, suggesting that budget cuts in specific areas may indirectly impact enrollment rates. Therefore, the findings of this study provide valuable insights for educational policymakers to adopt evidencebased strategies, ensuring more efficient planning and public policies aimed at improving the quality of public education.