Desenvolvimento de um score para análise de risco de evasão de estudantes do ensino superior baseado em aprendizado de máquina

Abstract

Dropping out of Higher Education contributes to great social, economic and academic loss. Among the main reasons for dropping out are the student’s difficulty in following the content, the structure proposed by the course and the lack of financial resources. In recent years, several studies have emerged to try to identify groups of students at risk of dropping out, either by identifying the factors that can contribute to dropout, or by creating classifiers based on Machine Learning. However, researches focus essentially on categorical indicators, that is, with binary results, which denote that the student is or is not in the risk group. This type of analysis is important, however, it does not show the variation in the student’s performance during their academic life, in addition to not offering a score within a performance score. Differently, this project use Machine Learning techniques in the creation of a Score, in order to provide a thermometer to analyze how close the student is or not to the dropout group. Preliminary results are promising, because when using 𝐾𝑁𝑁 to create the Score, it was possible to develop a Score with the best result of hyperparameters found in the experiments.