Exploring Feature Reduction for Dropout Predicting in Higher Education in Brazil

Dropout within the higher education system is a prevalent and intricate phenomenon characterized by a multitude of reasons that can differ significantly from one context to another. Developing machine learning models and discerning the key features for diverse contexts poses a considerable challenge. In this paper, we propose a process based on feature selection to create and evaluate machine learning models for predicting dropout in the higher education system. The approach not only outlines the essential steps for model development in any context but also emphasizes the identification of the most critical features. We conducted a comprehensive study across five distinct contexts within Brazilian higher education, specifically focusing on face-to-face courses. Through this process, we identified the most important features for predicting dropout. The results highlight that the correlation between a student's enrollment duration and the percentage of the course completed emerges as the primary predictor of dropout. However, we noticed the fundamental role of context in predicting dropout. Moreover, in all scenarios explored, it was possible to create more accurate models with a reduced set of features compared to the original models.