Submitted by Sara Ribeiro (sara.ribeiro@ucb.br) on 2017-11-25T11:49:15Z No. of bitstreams: 1 WeslleyRodriguesSepulvidaDissertacao2016.pdf: 1633691 bytes, checksum: 6449c77ed21fbc52e7c304fd7364796d (MD5) / Approved for entry into archive by Sara Ribeiro (sara.ribeiro@ucb.br) on 2017-11-25T11:49:41Z (GMT) No. of bitstreams: 1 WeslleyRodriguesSepulvidaDissertacao2016.pdf: 1633691 bytes, checksum: 6449c77ed21fbc52e7c304fd7364796d (MD5) / Made available in DSpace on 2017-11-25T11:49:41Z (GMT). No. of bitstreams: 1 WeslleyRodriguesSepulvidaDissertacao2016.pdf: 1633691 bytes, checksum: 6449c77ed21fbc52e7c304fd7364796d (MD5) Previous issue date: 2016-08-29 / Distance Education has grown along the years. Several educational institutions have been offering courses of internal improvement and qualification, as well as extension, undergraduate and postgraduate courses. The increase in the number of courses and the significant raise in the number of students result in new challenges to the educational institutions. The high dropout rates, common in Distance Education courses, is one of the crucial problems the institutions have to deal with. In this context, Data Mining is one of the main approaches for the development of predictive methods of evasion. The present paper aims an analysis of the Distance Education evasion in a traditional Midwest Brazilian University. The study intends to identify the behavior of the students who have dropped out undergraduate courses, in order to provide subsidies for the subjects involved in the teaching-learning process. As a preventive solution to the evasion issue, it is understood that proper communication with the students who are about to evade can lead to changes that contribute to minimize the problem. This study is presented in four parts: (i) literature review based on theoretical framework on Distance Education evasion, Virtual Learning Environments, Knowledge Discovery in Database (KDD) and Education Data Mining; (ii) an analysis of the Virtual Learning Environment institution database applying KDD techniques to identify the course abandonment behavior; (iii) development and validation of a model for predictive identification of students prone to evade; (iv) management actions to mitigate the problem. The results show that, when applying KDD to the variant data in time, 30 days after the beginning of the classes, it is possible to significantly predict evasion. From the results, an evasion prediction model was developed, as well as an evasion combat model. / A educa????o a dist??ncia (EAD) tem crescido nos ??ltimos anos. V??rias institui????es de ensino t??m ofertado cursos que v??o desde aperfei??oamentos e capacita????es internas at?? cursos de extens??o, gradua????o e p??s-gradua????o. Com o crescimento da oferta de cursos e o aumento significativo dos estudantes, as institui????es educacionais se colocam frente a novos desafios, entre eles o combate das altas taxas de evas??o, comum em cursos na modalidade EAD. Nesse sentido, a Minera????o de Dados ?? uma das abordagens que vem sendo explorada para o desenvolvimento de m??todos preditivos de evas??o. O presente trabalho prop??e uma an??lise da evas??o no contexto da EAD de uma tradicional universidade do Centro Oeste. O estudo busca identificar comportamentos dos estudantes que abandonaram cursos de gradua????o nessa modalidade, de maneira a fornecer subs??dios preditivos para os atores envolvidos no processo de ensino-aprendizagem de forma a apoiar a tomada de decis??es preventivas a respeito da evas??o. Parte-se do pressuposto de que interven????es junto aos estudantes propensos a evadir podem acarretar mudan??a no comportamento via Ambiente Virtual de Aprendizagem (AVA) que contribuem para minimizar a evas??o. Este trabalho ?? dividido em quatro partes: (i) uma revis??o da literatura, para embasamento te??rico, sobre evas??o no ??mbito da EAD, ambientes virtuais de aprendizagem (AVA), knowledge Discovery in Database (KDD) e Education Data Mining; (ii) an??lise da base de dados do AVA utilizado pela institui????o aplicando t??cnicas de KDD, para identifica????o do comportamento de abandono do curso; (iii) gera????o e valida????o de um modelo para identifica????o preditiva de estudantes propensos a evadir; (iv) A????es Gerenciais propostas para mitiga????o do problema. Os resultados mostram que ao aplicar o KDD nos dados variantes no tempo, com 30 dias ap??s o in??cio das aulas, ?? poss??vel predizer evas??o com precis??o significativa. A partir dos resultados obtidos, gerou-se um modelo de predi????o de evas??o bem como um modelo de tomada de decis??o e a????es de combate ?? evas??o.