Logistic regression and learning personalization in higher education

Resumo

This study focuses on the early prediction of academic performance in higher education students to personalize the learning process and enable timely interventions. Logistic Regression, widely used for its interpretability and effectiveness, serves as a starting point to assess its validity and utility in the context of higher education. A dataset of 10,184 students from the Universidad Autónoma de Baja California was analyzed. Three variable configurations (Basic, Complete, and Exam) and three classification algorithms (Logistic Regression, Naive Bayes, and Decision Tree) were compared using five-fold cross-validation and random sampling (90% training, 10% testing). Accuracy, Recall, F1 Score, and AUC-ROC were employed as evaluation metrics. Logistic Regression (Basic configuration) achieved the best metrics, yielding a Recall near 0.88 and an AUC-ROC around 0.72–0.76, outperforming Naive Bayes and Decision Tree. High school GPA emerged as the most influential variable, followed by Writing scores. These findings highlight the potential of Logistic Regression for early risk detection and learning personalization, although further investigation is warranted to address predictive fairness and incorporate socio-emotional factors that ensure a more inclusive and effective educational approach.