ESTUDO DA EVASÃO EM ALUNOS DE GRADUAÇÃO POR MEIO DE MINERAÇÃO DE DADOS

Abstract

University evasion is a major problem for the educational management.The most efficient way to address evasion is to predict and thereby take steps to prevent such action from occurring and for such anticipation to occur, analysis of student data patterns that evade is used. But for a more accurate analysis, tools are needed to help managers more accurately and efficiently and that is why data mining is used. This paper performed a bibliographical review analyzing the methods used in the evaluation of this prediction, algorithms used, software used and performance rates of several works of the last ten years. Subsequently, tests were applied to the database of students from Federal Institutes of the year 2017 available on the Nilo Peçanha Platform, using 13 Weka open source classification algorithms to evaluate which algorithm would perform best and which would be the most recommended to be used in an application that predicted such evasion. It was found that J48 was the most efficient, with others with very close performance and that the Paired Sample T-Test was fundamental for a more accurate comparison of the algorithms, since individual tests can be affected by external factors and influence the results, besides the fact that it is a test with higher statistical value.