Mineração de dados para identificação de fatores de reprovação no ensino superior

• Autor(a) principal: Dutra, Bruna Luiza
• Data de Publicação: 2019
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/28415
• topic: Evasão; Mineração de dados; Classificação; Fatores de reprovação; Dropout; Data mining; Data classification; Retention factors; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Brazil and several countries are facing the problem of students dropout in the universities. Particularly, dropout rates of computer science related courses are among the highest ones, with approximately one of three students receiving the diploma. For example, recent studies showed that some classes of the Information Systems course at Federal University of Uberlândia achieved dropout rates greater than 70%. Among the several reasons for students dropout, one of the most cited in the literature is students retention. In this sense, Data Mining methods can be adopted to find indicators that contribute to student dropout or retention, which can be a relevant tool to alleviate this problem. By considering a data set of technical, social and economic features of students from the Information Systems course at Monte Carmelo Campus, this work aims at evaluating data classification techniques in the problem of predicting the students final performance regarding the discipline of Introduction to Computer Programming. The present study also seeks to identify the factors that most contribute to a student's poor performance. The experiments were performed with the Weka tool. A total of six classification techniques were considered, in addition to two database situations: with and without missing attributes. To compare the results we adopted two metrics: accuracy and F1. Statistical tests attested the good results of the JRIP rule induction model in comparison with the other techniques. Among the indicators that stand out are: family per capita income, average math grade in high school, distance from parents city to the university city, distance from where student lives to the campus, type of school attended in high school and with who the student resides. It is expected that such indicators can assist course managers in the early identification of students who are most likely to fail and also in the planning of actions to reduce dropout rates.