Identificação de padrões para a análise da evasão em cursos de graduação usando mineração de dados educacionais

Abstract

Educational data mining is a recent research area that is gaining popularity because of their potential for educational institutions. One of the challenges of these institutions is to reduce the course dropout. The dropout in higher education is a phenomenon in growth and has become the focus of concern for researchers from different areas. However, the avoidance features are poorly studied and there is a lack of information and identification of models of their motives. This research proposes a computational approach for identifying patterns to be used in the analysis of dropout students in undergraduate classroom courses, in order to assist decision-makers in educational institutions. The proposed method selects the best attributes for classification task, in which the classes “dropout” and “non-dropout” are considered, based on the feature subset selection and feature creation. The experiments were conducted with the undergraduate students’ data at the Federal University of Technology - Paraná, consolidated in a Data Warehouse, that allowed the dropout investigation between the years 1980 and 2014. In this research are discussed the most common problems that occur in educational data mining, such as feature subset selection, unbalanced data, outliers and overfitting. The experimental results show the most relevant attributes to dropout prediction, indicating the contribution of the feature creation in the data mining task, allowing with these inferences to support the decision-making by educational managers located in strategic, tactical and operational levels.