Investigação da evasão estudantil por meio da mineração de dados e aprendizagem de máquina

Abstract

A student dropout is a recurring problem in both schools and universities, posing detrimental effects on students’ learning experiences and causing financial losses for institutions, whether public or private. Studies employing data mining (DM) and machine learning (ML) techniques to investigate educational issues are on the rise, and student dropout is one such problem. Through these techniques, it becomes possible to identify patterns in individuals or groups that may be prone to discontinuing their studies. Predictions and insights gained through DM and ML techniques can aid in reducing student dropout rates by providing a better understanding of the phenomenon, thereby informing decision-making and problem resolution. This study aims to investigate university dropout using data from the Higher Education Census obtained from the INEP portal. It analyzes the classification of university student dropout using classical algorithms in building classification models. The investigation reveals that, based on the 2016 Census, 30% of university students end up abandoning their enrolled courses or discontinuing higher education altogether. Additionally, it highlights that existing studies utilizing dropout classifiers lack a standardized set of attributes for constructing their models, making it challenging to apply these models across different institutions. To address these issues, the study employs INEP data in constructing classification models during the experiments. The best result, with an accuracy metric of 0.979, was achieved by the model based on the Decision Tree algorithm in the distance education (Educação a Distância - EAD) modality and public administrative category.