Técnicas de inteligência artificial para predição da evasão em cursos técnicos na rede federal de educação profissional, científica e tecnológica

Abstract

School dropout is widely recognized as one of the greatest educational challenges in all countries. Despite the relevance of the topic, there are few studies that identify the factors from a quantitative perspective. This is mainly due to the lack of longitudinal data, which is essential for an adequate assessment of the factors that lead to dropout throughout a student's academic career. Understanding the causes of dropout and its characteristics can provide valuable insights for the development of strategies and policies to reduce dropout in technical courses offered by the Federal Network of Professional, Scientific, and Technological Education in Brazil. The objective of this project is to analyze, using data extracted from the National System of Information on Professional and Technological Education and disclosed by the Nilo Peçanha Platform, the factors that lead students in technical courses to abandon their studies, through the use of Artificial Intelligence techniques. The research is quantitative in nature, using bibliographic and documentary research techniques. The analysis and interpretation of the results will provide input for the development of measures capable of addressing this significant issue, which has such a major impact on students' education and technical training: dropout rates. This work is divided into three articles. The first presents the machine learning techniques used to predict dropout rates in technical courses in the Federal Network. The second article presents the analysis and interpretation of dropout results using Shapley Additive Explanations (SHAP). Finally, the third article presents the PrevIA platform, a simulator developed with artificial intelligence to provide the probability of student dropout based on the insertion of new data related to their personal and sociodemographic characteristics and the intended course. The results of the computational modeling show satisfactory performance of the model, with a recall rate of 69% accuracy and an area under the ROC curve (AUC) of 78%, demonstrating the approach's ability to support decision-making processes that lead to dropout and, consequently, to formulate better student assistance policies to reduce the effects of this phenomenon, which causes social, academic, and financial losses for professional education in the country.