Previsão de evasão em cursos de ensino superior através de aprendizado de máquina associado à análise de disciplinas aprovadas

Abstract

The scholar dropout on higher education context reaches dangerous levels. As studies DAVOK; BERNARD (2016), about 50% of exact science students don’t finish their respective courses. To help finding a useful solution, machine learning algorithms has been used more frequently in order to search for education patterns and to identify students with dropout potential. However, finding a attribute combination which is efficient for the training step is not a simple assignment. Moreover, the big rise of system computational powerful made possible the development of better data mining techniques, and from the increasingly amount of generated and processed data. Thereby, this work’s main purpose is to make an attribute combination based on information about approved disciplines from a student, with a classifier technique, to achieve expressive accuracy levels on students dropout potential. A good dropout student hit level is primordial to improve the searches for effective measures on dropout control, so it can be used for the student pedagogical monitoring.