Prediction of academic dropout in higher education - the case of face-to-face undergraduate courses at the University of Brasília

Abstract

Researchers have been investigating attrition in higher education, identifying two categories - students who leave the university and those who abandon higher education altogether, negatively affecting institutions, students, and society. Since 1995, in Brazil, the creation of ANDIFES has led to more frequent studies on graduation, retention, and attrition in Brazilian universities, focusing on institutional attrition, characterized by a student's departure from their original course. The University of Brasília (UnB) has implemented strategies to increase student retention in undergraduate courses. This work aimed to develop and test an analysis model to predict attrition in face-to-face courses, through a Systematic Literature Review to identify impact factors and define indicators extracted from UnB's academic systems. The model, named MAGRA, combines indicators with machine learning algorithms to identify students at risk of dropping out. Tests conducted at the Faculdade de Ciências e Tecnologias em Engenharia (FCTE) and UnB indicated that enrollment frequency in subjects could predict completion difficulties. The research suggests that to improve the early identification of at-risk students, adjustments in feedback mechanisms, the inclusion of new systems, improvement in data quality, and adjustments in algorithm parameters are necessary.