Aplicação de redes neurais RBF e MLP na análise de evasão discente do curso de Sistemas de Informação da UFRRJ

Abstract

The school dropout is a complex problem, which gathering various factors that influence the decision of the student to stay or not in the course. The identification of the characteristics involved in this process turned it possible to analyse the profile of the students prone to evasion, and thereby to take effective action. According to the literature, courses in Information Technology area have shown high dropout rates. However, the course of Information Systems of UFRRJ has shown alarming rates. Among the many studies in the literature to predict dropout in higher education in Brazil, the use of computational techniques has aroused the interest of researchers, precisely Artificial Neural Networks ARTMAP Fuzzy, Radial Basis Function (RBF) and Bayesian networks, which has been showing good results in predicting student dropout in higher education institutions. The objective of this dissertation is to propose an intelligent system based on Multilayer Perceptron (MLP) and Radial Basis Function (RBF) in student evasion analysis of the course of Bachelor of Information Systems UFRRJ. The used data set corresponds to 148 samples of students enrolled in the course. For such a study was considered evasion to the 3rd period , where the following variables were used: gender, school origin, performance in the specific subjects of the 1st period, distance, time between high school and higher education, performance in ENEM (National Secondary Education Examination), civil status and situation regarding evasion. It were performed 100 simulations, being selected the top 25. The results obtained by the RBF and MLP networks prove to be promising in the course Student dropout analysis. Since the RBF network performed better (accuracy = 91.03%) than MLP model, with a significant difference (p value<0.05) between the results of the models.