<![CDATA[The rise in academic dropout rates at universities across Brazil is notable, especially in the post-COVID-19 pandemic period. At the Federal University of Paraná (UFPR), the problem escalated after 2019, reaching peaks in 2021 and 2022, underscoring the urgency for targeted intervention. This study aims to compare different Machine Learning methods - Logistic Regression (LR), Decision Trees (DT), Random Forest (RF), Stochastic Gradient Boosting (GBM), Support Vector Machines (SVM), and Neural Networks (NN) - using institutional data in undergraduate courses from UFPR (2009-2024). The fitted models are based on academic performance and course-related variables, specifically the Academic Performance Index (IRA), the Proportion of Failed Courses (PropRep), the course's academic type (Degree), and its area of knowledge (THE Area). Results showed that while all models achieved similar predictive performance (MCC ~ 0.70 | Precision > 0.90), in general, the Logistic Regression model offers high interpretability and easiness to implementation. The LR model is proposed as the most robust and transparent predictive tool for the university's management.]]>
Copyrights © 2026
