Previsão de evasão universitária com aprendizado de máquina

Abstract

Student dropout is a problem that affects higher education institutions around the world, having negative impacts on both students and institutions, whether public or private. It is essential that institutions have tools that help them control evasion, providing managers with an understanding of the educational expectations of students entering higher education, in order to improve understanding of this phenomenon. Recent studies on university dropout prediction using machine learning represent a significant advance in the area of education. By employing the Cross-Validation Technique (K-Fold) along with a variety of classification algorithms such as decision trees, logistic regression, random forest, and support vector machines, among others. This work seeks to understand and anticipate dropout patterns among students. This approach not only identifies risk factors for dropout, but also provides valuable information for educational institutions in developing proactive student retention strategies. By accurately predicting the likelihood of a student dropping out of their studies, universities can intervene early, offering personalized support and additional resources to help students overcome academic and personal challenges. To achieve this, in relation to the student recovery model, LogisticRegression techniques , GradientBoosting and XG Boost obtained similar and promising results, above 90% for graduate F1-Score and dropout F1-score close to 89%. As for the cases of interpretable algorithms, model for student dismissal, the best results were for the Random Forest and Decision Tree models with values of 91% for Graduate F1-Score, 84% for dropout F1-score. This work represents a significant contribution to improving the quality and effectiveness of educational programs, promoting the retention and success of university students.