| 要約: | Student attrition and eventual dropout are problems that affect many universities around the world. In Brazil, there are no official statistics to monitor them. In this work, data
mining techniques were used in order to unveil a profile of dropout students from the Brazilian higher education. The data from Brazil’s higher education census, CES, and its
nation-wide high school exam, ENEM, were used to create multiple classification models, ranging from five different classification methods and three separate dropout definitions. The experiments were conducted on UnBs students data. Among the classification methods, CART showed a subtle lead, performance wise. It obtained a sensibility score of around 84% when the dropout definition was focused on the students major. On the other two dropout definitions, there wasnt a statistically significant difference between the tested methods. The main characteristics for the dropout students unveiled by the generated models were: to enter the university in the first semester, attend to more than one institution, obtain higher than average grades on the high school examinations and finally, having graduated from high school when taking the ENEM exam. Furthermore, a R package was developed in order to train new classifiers for dropout. It can be used to determine, in a given database, which students are more likely to dropout.
|
|---|