Prediction and Explainability of School Dropout in Peru Using Machine Learning and Explainable Artificial Intelligence

Barboza A.; Huamani J.; Santisteban J.; Maculan N.; Mauricio D. · IEEE Access (2026) · DOI: 10.1109/access.2026.3675564

Resumen

School dropout represents a critical challenge for educational systems in developing countries. This study proposes an integral framework that combines machine learning techniques and explainable artificial intelligence to predict and explain school dropout in the Peruvian context. The proposal incorporates key variables in four dimensions: personal, family, sociodemographic, and academic. The study implements and compares five machine learning algorithms (Random Forest, Support Vector Machine, Gradient Boosting, K-Nearest Neighbors, and XGBoost) along with SHapley Additive exPlanations (SHAP) for explainability, integrated into a web application called EduPredict. The experiments conducted on a dataset of 1,456 records from a public school in Metropolitan Lima, using 20 predictive factors, demonstrate exceptional performance, with Gradient Boosting achieving an accuracy of 97.27% and precision of 99.28%. The explainability analysis using SHAP identified behavior and academic performance as the most determining factors, significantly contributing to the dropout risk. The validation with 10 educational professionals through structured surveys revealed high levels of usability, utility, and satisfaction, confirming the practical applicability of the system in real educational environments. The framework has potential to be adapted to other educational contexts in developing countries, offering a personalized approach to mitigate school dropout.

Instituciones: Universidad Nacional Mayor de San Marcos; Universidad Peruana de Ciencias Aplicadas