Técnicas de classificação para a predição da evasão universitária

Abstract

This research pursues to predict the classification of students regarding dropout through the application of Machine Learning techniques. For this, a theoretical grounding was made on the possible causes of student dropout and the obtaining of a set of data through the university system itself. First, in Experiment 1, the Decision Tree, kNN, and RNA technique were used, and then, in Experiment 2, the kNN and RNA technique was used, but with reduced attributes instead, selected by the Decision Tree in Experiment 1. It was possible to conclude that the most appropriate technique for predicting student dropout was the RNA technique, with a learning rate of 0.1, with selection of attributes, which obtained the best performance presenting accuracy of 92,3%, precision of 89.8%, and recall of 82,6 %. It was also possible to verify that the main factors that can influence dropout are issues related to students's academic performance.