Mineração de dados educacionais: uma análise sobre as variáveis que influenciam na evasão escolar

Abstract

The school evasion is a big problem that concerns the Brazilian educational scenario. This paper presents some factors that cause this concern, and a solution to part of the problem is presented by using Educational Data Mining. The objective is to identify variables related to school dropout of elementary school students from Amazonas State schools and to create a model for predicting the probability of dropout using machine learning methods. The methodology was based on the CRISP-DM model, widely used for data mining. Three prediction learning models were trained and evaluated: Logistic Regression, Decision Trees and Random Forests. Among the variables used, it is noteworthy that most of them correspond to school location data, school equipment and student age.