Aplicação de técnicas de aprendizado de máquina para classificar alunos de cursos de idiomas com relação à possibilidade de evasão

Abstract

The increase in competitiveness in the labor market has led people to seek new skills and knowledge, among which are the language courses. However, the courses have been suffering from high dropout rates, which is caused by multiple factors, which have complex relationships that make difficult the elaboration of classification models. In this context, the current research proposes the use of Machine Learning techniques capable of analyzing large and complex databases, with the purpose of identifying in advance a student who is prone to evade the language course, thus enabling measures to be taken reduce the rate of evasion. In order to study this problem, a literature review was carried out on the factors that can cause student dropouts, and then a questionnaire was developed using Google Forms. The questionnaires were pre-processed and 7 machine learning techniques were used in the study of two classification models, each with two different configurations. The first model aimed to predict whether the student was attending, whether he had escaped or had completed the language course, while the second was intended to predict only whether the student evaded or not. The results were satisfactory, with emphasis on the techniques of Support Vector Machines and Random Forests, which obtained a maximum accuracy of 91% and 88%, respectively.