Predição do desempenho no ENADE dos discentes de computação

Abstract

The evaluation of higher education is important to ensure its quality, in addition to being of interest to students, teachers, governments and society. The National Higher Education Assessment System (SINAES) is an educational assessment network with several assessment methods, one of which is the National Student Performance Examination (ENADE), which is organized by the National Institute for Educational Studies and Research Anísio Teixeira (INEP). ENADE evaluates undergraduate course students who are in the final year of the course or who have completed at least 80% of the course workload. This exam has an evaluation cycle of three years, evaluating a set of different courses in each of these years, with courses in the computing area being evaluated in Year II. The results of assessments and complementary questionnaires are made available as open data by INEP, serving as a rich source for analysis and the discovery of new knowledge. Although the data are available, scientific research that explores their relationships and evolutions is still restricted, especially with regard to the identification of factors that help decision-making by managers of Higher Education Institutions. With this in mind, the objective of the present work is to use the Discovery of Knowledge in Database method together with classification techniques to predict the performance of computing students participating in ENADE based on socioeconomic factors, seeking to discover the best model. of classification and what is the best cut-off point to generate classes of the participants’ performance. In this work, five classification models were evaluated: Decision Tree, K-nearest neighbors, Support Vector Machine, Random Forest and Logistic Regression; in addition to testing five ways of classifying participants according to performance. As a result, it was found that the best combination for classifying participants’ performance is to use the Support Vector Machine model, adopting the median as a way of separating the classes of participants. The result obtained contributes to the understanding of how ENADE data can be used to predict the performance of participants, in the definition of Machine Learning models suitable for this context, in addition to the discovery of how the way participants’ performance is divided impacts the performance of classification models and which one is the most suitable.