Utilização de aprendizado de máquina na predição de desempenho acadêmico

Abstract

This study investigates the application of machine learning models to predict academic performance, focusing on logistic, polynomial, and linear regression techniques. The main goal is to identify patterns in variables such as study time, attendance, and parental involvement to forecast grade categories (e.g., "A", "B" "C"). The methodology involved practical implementation in Python using libraries like Pandas and Scikit-learn, with cross-validation and hyperparameter tuning (e.g., 1/L2 regularization). Exploratory analyses revealed key insights, such as performance decline after 10 absences and the positive impact of tutoring (+0.3 points on average). Models achieved up to 77.16\% accuracy (logistic regression) and 71.46\% $R^2$ (polynomial regression), highlighting their potential for personalized educational interventions. The study emphasizes the value of interpretable and accessible tools, aligning with global demands for educational equity.