Predicción del riesgo académico del estudiante universitario aplicando algoritmo de ensemble y redes neuronales
Loading...
Código QR
Authors
Campomanes Murrugarra, Fanny
Contact Email
Abstract
El presente estudio tiene como objetivo identificar el algoritmo que tenga mejor precisión en predecir el riesgo académico de los estudiantes de la Universidad Nacional Agraria La Molina mediante la aplicación de diversos algoritmos de aprendizaje automático. Se evaluaron modelos de tipo ensemble, entre ellos Bagging, Random Forest y XGBoost además de CART y Redes Neuronales. La comparación del desempeño se realizó utilizando datos de prueba y validación cruzada. Los resultados muestran que las redes neuronales constituyen la opción más adecuada, destacando por su alto recall (0,9163 y 0,9117) y especificidad (0,8105 y 0,7819), así como por un AUC-ROC superior (0,8702 y 0,8567). Bagging y XGBoost se presentan como alternativas secundarias, aunque su menor especificidad podría limitar su aplicabilidad. En contraste, el modelo CART queda descartado debido a su rendimiento significativamente inferior. Por lo tanto, las redes neuronales ofrecen la mayor capacidad predictiva y resultan especialmente valiosas para la identificación temprana de estudiantes en situación de riesgo académico.
The present study aims to identify the algorithm with the best accuracy in predicting the academic risk of students at the National Agrarian University La Molina through the application of various machine learning algorithms. Ensemble-type models were evaluated, including Bagging, Random Forest, and XGBoost, in addition to CART and Neural Networks. Performance comparison was conducted using test data and cross validation. The results show that neural networks constitute the most suitable option, standing out for their high recall (0,9163 and 0,9117) and specificity (0,8105 and 0,7819), as well as for a superior AUC-ROC (0,8702 and 0,8567). Bagging and XGBoost are presented as secondary alternatives, although their lower specificity could limit their applicability. In contrast, the CART model is ruled out due to its significantly inferior performance. Therefore, neural networks offer the greatest predictive capacity and are especially valuable for the early identification of students in academic risk situations.
The present study aims to identify the algorithm with the best accuracy in predicting the academic risk of students at the National Agrarian University La Molina through the application of various machine learning algorithms. Ensemble-type models were evaluated, including Bagging, Random Forest, and XGBoost, in addition to CART and Neural Networks. Performance comparison was conducted using test data and cross validation. The results show that neural networks constitute the most suitable option, standing out for their high recall (0,9163 and 0,9117) and specificity (0,8105 and 0,7819), as well as for a superior AUC-ROC (0,8702 and 0,8567). Bagging and XGBoost are presented as secondary alternatives, although their lower specificity could limit their applicability. In contrast, the CART model is ruled out due to its significantly inferior performance. Therefore, neural networks offer the greatest predictive capacity and are especially valuable for the early identification of students in academic risk situations.
Description
Universidad Nacional Agraria La Molina. Escuela de Posgrado. Maestría en
Estadística Aplicada
Keywords
Deserción universitaria; Educación agraria; Educación pública; Programa de ordenador
Citation
Date
2026
Collections
Seleccionar año de consulta:
Licencia de uso

Excepto si se señala otra cosa, la licencia del ítem se describe como info:eu-repo/semantics/openAccess

