Técnicas de aprendizado de máquina para descoberta de conhecimento sobre dados abertos do ensino superior público brasileiro

Abstract

This work deals with the use of techniques from the methods of Knowledge Discovery in Databases — KDD — and Cross Industry Standard Process for Data Mining — CRISPDM — on educational databases made available by the Instituto Nacional de Estudos e Pesquisas Educacionais Anísio Teixeira (National Institute of Educational Studies and Research Anísio Teixeira) — INEP — aiming to the discovery of knowledge that allows the identification, as well as the understanding, of the context in the formation of students from public Brazilian higher education institutions. Three data mining scenarios are observed, considering Supervised Machine Learning and Unsupervised Machine Learning methods, covering data classification, grouping and association experiments. The first scenario, which includes data from graduates of bachelor’s and licentiate’s undergraduate courses, aims to predict the approximate length of stay at graduation, considering the students’ socioeconomic information, through 16 classification models built using Decision Tree, Random Forest, XGBoost and Multilayer Perceptron Neural Network algorithms. XGBoost models had the best results in all experiments. In turn, the second scenario uses the Kmeans algorithm to perform a grouping of public higher education institutions that, based on the analysis of four groups obtained by considering information on expenses, numbers of professors and technicians, location and administrative category, among others, made it possible to identify similarities and dissimilarities between the institutions. The groups, in addition to data used in the first scenario, which include information about the students, such as age group, length of stay at graduation and form of admission to graduation, noting whether this occurred through affirmative action or social inclusion policies, among others, they are considered in the experiments of the third scenario, using the Apriori algorithm, for the generation of association rules that can support the discovery of knowledge in the context of Brazilian public higher education.