Predição da evasão acadêmica aplicando análise temporal

• Ano de defesa: 2021
• Autor(a) principal: Vieira, Raphael dos Santos Guedes
• Orientador(a): Nascimento, Hugo Alexandre Dantas do
• Banca de defesa: Nascimento, Hugo Alexandre Dantas do, Monsueto, Sandro Eduardo, Ferreira, Deller James, Mello, Rafael Ferreira Leite de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Ciência da Computação (INF)
• Departamento: Instituto de Informática - INF (RG)
• País: Brasil
• Palavras-chave em Português: Predição da evasão acadêmica; Divisão temporal; Mineração de dados educacionais; Temporal learning analytics
• Palavras-chave em Inglês: Academic dropout prediction; Temporal split; Educational data mining; Temporal learning analytics
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/11793

Academic dropout in higher education is a recurring problem in the daily life of public and private, national and international educational institutions. When a student drops out, this generates consequences in the social, professional, and financial domains both for him/her and for the academic environment where he/she is inserted, which reflects on the national development. Computational methods to assist in predicting cases are an important tool for dealing with this phenomenon. However, student progress is an activity that takes place over time, turning dropout prediction into a temporal problem, and this aspect has been little explored in the literature. The present work aims to contribute to filling this gap, by adapting and expanding a temporal predictive approach that combines unsupervised and supervised learning to predict dropout in the context of Brazilian federal universities. Furthermore, a new method of data segmentation in training and testing is experimented that seeks to reflect, more effectively, the real and temporal scenario of dropout prediction. The approach is tested on a set of supervised machine learning algorithms and evaluated using data extracted from two academic units of the Universidade Federal de Goiás (UFG), between the years 2009 and 2020. It is observed, in the end, that the temporal approach of both the method and the data segmentation provide more realistic results.