Data mining e data analytics para apoio à gestão estratégica e mitigação da evasão escolar

• Ano de defesa: 2019
• Autor(a) principal: Vasconcelos, Nathanael Oliveira
• Orientador(a): Rodrigues Júnior, Methanias Colaço
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Programa de Pós-Graduação: Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Computação; Evasão; Mineração de dados educacionais; Algoritmos de classificação; Experimentação
• Palavras-chave em Inglês: School dropout; Educational data mining; Classification algorithms; Experimentation
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://ri.ufs.br/jspui/handle/riufs/16599

Context: Dropout is certainly one of the major problems that plague educational institutions in general, since, as a result of student dropout, they are social, academic and economic assets. A search for their actions has been the subject of much educational work and research around the world. In the practical field, the various American high school associations had their guidelines focused on the dropout rate control, however, in Brazil, there is still little work done in this area of research. As a result, there is the ability to increase understanding of the problem and its causes by adopting more effective measures to identify and understand the key factors that may contribute to student failure. Objective: This work had the purpose of conducting experiments of more advanced data mining algorithms in the area of education, aiming to improve the educational context of the high school dropout of two federal institutions, as well as to implement a method of using the best model, which supports the decision support process and the school dropout mitigation. Method: Two in vivo controlled experiments were developed and performed to compare the selected classifiers. Then, a case study with interface created to apply the algorithm that obtained the best result was performed. Results: The results showed the differences between the algorithms used and, despite the SVM, had a higher average of the accumulation metrics, statistically, after a meta-analysis of the experiments, the algorithms MLP and Random forest, respectively, obtained similar accuracy results (85.38 %, 84.40 % and 84.13 %). For F-measure, a statistical significance was equal only for the MLP (84.42 %, 83.44 %). Conclusions: This dissertation exposes the need to increase the admission of measures to identify and understand the main factors that may contribute to students’ failure. The two experimental tests were examined and introduced by the success criteria, being the SVM, the selected differential, selected to be applied in a case study of one of the planning items. Federal University of Sergipe - UFS.