Combinação linear de classificadores para análise de risco na evasão da UFC

Abstract

With the change from the selective process from the entrance exam to the ENEM / SiSU at the Federal University of Ceará - UFC, the phenomenon of student evasion became more precocious, occurring in the majority of the first two semesters of the course. This phenomenon causes considerable damage to the institutions as well as to society as a whole. In this way, a study on this problem and the elaboration of mechanisms to perceive the possibility of avoidance as quickly as possible is necessary. One of the possible actions is the identification of the students subject to evasion based on some evaluative metric, denoted here by classifier. For the selection of good classifiers, there are different measures to evaluate its efficiency, such as accuracy, sensitivity, specificity and the ROC (receiver operation characteristic) curve. This is quite applied in several areas of knowledge. From this, several other measures are obtained, such as the AUC (area under curve) and pAUC (partial area under curve), these measures are based on a considerable theoretical repertoire. The objective of the present work is to approach techniques aimed at combining different classifiers in order to form a better discriminative capacity. Applying this tool in practical data related to the students entering the UFC, in the courses of the Technology Center, to verify the ability of these methods in the perception of the evasion risk of these students with the use of basic grades of ENEM.