Mineração de Dados Educacionais para a predição da evasão de alunos do ensino superior: Estudo de caso na Universidade Federal da Fronteira Sul - Campus Realeza

The reduction of high dropout rates in higher education institutions has proven to be a challenge for educational administrators. In order to assist in mitigating the high rates of abandonment, several studies have utilized Educational Data Mining techniques to discover patterns that effectively ind...

Descripción completa

Detalles Bibliográficos
Autor: Falcão, Adair Perdomo
Tipo de recurso: tesis de maestría
Estado:Versión publicada
Fecha de publicación:2023
País:Brasil
Institución:Universidade Estadual do Oeste do Paraná (UNIOESTE)
Repositorio:Biblioteca Digital de Teses e Dissertações do UNIOESTE
Idioma:portugués
OAI Identifier:oai:tede.unioeste.br:tede/6934
Acceso en línea:https://tede.unioeste.br/handle/tede/6934
Access Level:acceso abierto
Palabra clave:Árvores de decisão
Classificação
Evasão
Mineração de dados educacionais
Decision Tree
Classification
Dropout
Educational Data Minin
CIÊNCIA DA COMPUTAÇÃO
Descripción
Sumario:The reduction of high dropout rates in higher education institutions has proven to be a challenge for educational administrators. In order to assist in mitigating the high rates of abandonment, several studies have utilized Educational Data Mining techniques to discover patterns that effectively indicate students with potential for dropout. Considering that the dropout rate is higher in the first year and that the majority occurs by the third semester, anticipating this risk gains importance, as predicting dropout as early as possible allows for the planning and execution of preventive actions more effectively. In this context, this exploratory and explanatory research proposes a predictive model of dropout based on the analysis of pre-university data, using educational data mining techniques and cost-sensitive classification, applied to data from 1086 incoming students in undergraduate courses at the Realeza Campus of the Federal University of Fronteira Sul, from 2014 to 2018. Through the J48 decision tree inductive algorithm, experiments were conducted on 18 databases extracted from the collected data. The models obtained achieved an average overall accuracy of 70%. Most of them obtained true positive rates above 80% and false negative rates below 20%, demonstrating a satisfactory ability to identify students prone to dropout in advance. The analyses performed emphasize the importance of taking into account the specificities of each group of students when developing prevention strategies and policies. The promising results suggest that it is feasible to predict student dropout based solely on pre-university data.