Predicción de la deserción estudiantil en la Universidad Tecnológica de Pereira mediante la implementación de modelos de machine learning.
Cargando...
Fecha
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Contexto. La Vicerrectoría de Responsabilidad Social y Bienestar Universitario de la Universidad Tecnológica de Pereira (UTP) lidera un plan de acompañamiento integral que brinda apoyo a los jóvenes desde el momento en que se inscriben en la universidad. Según el Observatorio Social de la UTP, se han identificado condiciones específicas que impiden que algunos jóvenes se integren a la vida universitaria o, si ya están dentro del sistema, que deserten. Sin embargo, hasta el momento, en la UTP no se ha llevado a cabo un estudio estadístico que analice las diferentes condiciones socioeconómicas de los estudiantes y con la permanencia o deserción. Propósito. Desarrollar un modelo predictivo para determinar la deserción o permanencia escolar de los estudiantes que ingresan a la UTP; este modelo se basará en un análisis de las características socioeconómicas de los estudiantes que ingresan desde primer semestre. Metodología. Se entrenaron 15 modelos de clasificación de machine learning con datos de 13,000 estudiantes que ingresaron entre 2015 y 2019. Resultados. El modelo de Gradient Boosting Classifier (GBC) obtuvo el mejor puntaje de AUC (AUC=0.77 con un umbral de 0.50). Sin embargo, se optó por elegir el modelo de Regresión Logística (RL), que a pesar de obtener un AUC 3 puntos porcentuales menos que el GBC (AUC=0.74 con un umbral de 0.70) puede ser de mayor utilidad por su facilidad de aplicación. Conclusiones. Dada la sencillez de interpretación del modelo de Regresión Logística, se considera que este podría ser especialmente útil para predecir la deserción escolar. Mediante la exploración y ajuste de sus parámetros, este modelo puede identificar a los estudiantes en riesgo de abandonar los estudios desde su ingreso. La implementación de esta herramienta en el Programa de Apoyo Integral (PAI) permitirá la detección temprana de deserción, facilitando el diseño y la aplicación de estrategias de retención que aumenten significativamente la probabilidad de éxito. Esto no solo mejorará la retención estudiantil, sino que también optimizará los recursos destinados a la educación.
Context. The Vice-Rector for Social Responsibility and University Welfare of the Technological University of Pereira (UTP) leads a comprehensive support plan that provides support to young people from the moment they enroll in the university. According to the UTP Social Observatory, specific conditions have been identified that prevent some young people from integrating into university life or, if they are already within the system, from dropping out. However, to date, at the UTP no statistical study has been carried out that analyzes the different socioeconomic conditions of the students and their permanence or dropout. Purpose. Develop a predictive model to determine the dropout or permanence of students who enter the UTP; This model will be based on an analysis of the socioeconomic characteristics of students who enter from the first semester. Methodology. 5 machine learning classification models were trained with data from 13,000 students who enrolled between 2015 and 2019. Results.The Gradient Boosting Classifier (GBC) model obtained the best AUC score (AUC=0.77 with a threshold of 0.50). However, it was decided to choose the Logistic Regression (RL) model, which despite obtaining an AUC 3 percentage points less than the GBC (AUC=0.74 with a threshold of 0.70) may be more useful due to its ease of application. Conclusions. Due to the straightforward interpretation of the Logistic Regression model, it is considered particularly useful for predicting school dropout. By exploring and adjusting its parameters, this model can identify students at risk of dropping out as soon as they enter. Implementing this tool in the Comprehensive Support Program (PAI) will enable early detection of potential dropouts, facilitating the design and implementation of retention strategies that significantly enhance the likelihood of success.
Context. The Vice-Rector for Social Responsibility and University Welfare of the Technological University of Pereira (UTP) leads a comprehensive support plan that provides support to young people from the moment they enroll in the university. According to the UTP Social Observatory, specific conditions have been identified that prevent some young people from integrating into university life or, if they are already within the system, from dropping out. However, to date, at the UTP no statistical study has been carried out that analyzes the different socioeconomic conditions of the students and their permanence or dropout. Purpose. Develop a predictive model to determine the dropout or permanence of students who enter the UTP; This model will be based on an analysis of the socioeconomic characteristics of students who enter from the first semester. Methodology. 5 machine learning classification models were trained with data from 13,000 students who enrolled between 2015 and 2019. Results.The Gradient Boosting Classifier (GBC) model obtained the best AUC score (AUC=0.77 with a threshold of 0.50). However, it was decided to choose the Logistic Regression (RL) model, which despite obtaining an AUC 3 percentage points less than the GBC (AUC=0.74 with a threshold of 0.70) may be more useful due to its ease of application. Conclusions. Due to the straightforward interpretation of the Logistic Regression model, it is considered particularly useful for predicting school dropout. By exploring and adjusting its parameters, this model can identify students at risk of dropping out as soon as they enter. Implementing this tool in the Comprehensive Support Program (PAI) will enable early detection of potential dropouts, facilitating the design and implementation of retention strategies that significantly enhance the likelihood of success.