Publicación: Predicción del rendimiento académico en las Matemáticas de la educación Secundaria mediante Redes Neuronales
Fecha
2020-10-14
Autores
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
Atribución-NoComercial-SinDerivadas 4.0 Internacional
info:eu-repo/semantics/openAccess
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Nacional de Educación a Distancia (España). Facultad de Ciencias. Departamento de Física Fundamental
Resumen
Las redes neuronales son sistemas de procesamiento de información, que se enmarcan
dentro del aprendizaje automático o machine learning. Por medio de esta
técnica se pueden obtener predicciones o resultados sobre problemas cuya resolución
implicaría un elaborado y complejo plan. La educación, es uno de los pilares fundamentales
de nuestra sociedad, y es permeable al empleo de las nuevas tecnologías,
como el machine learning para mejorar. Desde hace años se vienen utilizando técnicas
de data mining en este ámbito para analizar los datos que generan, por ejemplo,
los cursos universitarios online para obtener una predicción del rendimiento del
alumnado en este tipo de cursos.
En el presente trabajo se emplea esta técnica para la predicción del rendimiento
académico, con datos obtenidos en un entorno docente real, con alumnos de secundaria
de un colegio. Con una muestra de unos pocos registros, y un conjunto pequeño
de datos para cada uno (notas, identificador de grupo y medidas de atención a la
diversidad) se han entrenado una serie de redes neuronales multicapa feedforward
con el algoritmo backpropagation (propagación de errores hacia atrás), con el fin
de comprobar si estos sistemas son capaces de predecir el rendimiento académico.
También se quiere comprobar la calidad de esa predicción en función de la informaci
ón de la entrada, del número de unidades de capa oculta y si se mejora el resultado
complicando la con_guración de la red con una segunda capa oculta o combinando
las posibles salidas.
Con los resultados obtenidos se puede concluir que la predicción del rendimiento
académico en la etapa secundaria, con los datos obtenidos en un entorno docente
real, es posible. La precisión que ofrecen los resultados es razonablemente buena,
en muchos casos superior al 90 %, pero que disminuye a la hora de dar una nota
numérica. El incluir una estimación de las medidas de atención a la diversidad resulta
adecuado y refina en muchos casos los resultados obtenidos.
Artificial Neural Networks are information processing systems that belong to machine learning. Through this technique predictions or results can be obtained to solve problems that would require an overly complicated plan to reach a solution. Education, one of the fundamental cornerstones of our society, is especially receptive to the use of new technologies, such as machine learning. Over the last years, data mining techniques are being used in this context to analyze data generated by, for example, online university courses in order to obtain a performance prediction of its students. This technique is used in the present thesis to predict academic performance, using data obtained from a real educational context with secondary level students from a single school. Using a sample of a few students and a small set of data for each (grades, group identi_cation and attention to diversity measures) a series of feedforward multilayered neural networks have been trained with a backpropaga- tion algorithm in order to test if these systems are capable of predicting academic performance. Furthermore, a series of additional objectives are tested, such as: the prediction quality based on the input information, the number of hidden layer units, and if results can improve increasing the network's con_guration dificulty introducing a second hidden layer or combining the possible outputs. The conclusion that can be drawn given the results and obtained data from a real educational context, is that academic performance for secondary level students is possible. The accuracy o_ered by the results is reasonably good, reaching 90% in numerous cases but showing a decrease when using numerical marks. The inclusion of diversity attention measures proves to be adequate, re_ning the majority of the results.
Artificial Neural Networks are information processing systems that belong to machine learning. Through this technique predictions or results can be obtained to solve problems that would require an overly complicated plan to reach a solution. Education, one of the fundamental cornerstones of our society, is especially receptive to the use of new technologies, such as machine learning. Over the last years, data mining techniques are being used in this context to analyze data generated by, for example, online university courses in order to obtain a performance prediction of its students. This technique is used in the present thesis to predict academic performance, using data obtained from a real educational context with secondary level students from a single school. Using a sample of a few students and a small set of data for each (grades, group identi_cation and attention to diversity measures) a series of feedforward multilayered neural networks have been trained with a backpropaga- tion algorithm in order to test if these systems are capable of predicting academic performance. Furthermore, a series of additional objectives are tested, such as: the prediction quality based on the input information, the number of hidden layer units, and if results can improve increasing the network's con_guration dificulty introducing a second hidden layer or combining the possible outputs. The conclusion that can be drawn given the results and obtained data from a real educational context, is that academic performance for secondary level students is possible. The accuracy o_ered by the results is reasonably good, reaching 90% in numerous cases but showing a decrease when using numerical marks. The inclusion of diversity attention measures proves to be adequate, re_ning the majority of the results.
Descripción
Categorías UNESCO
Palabras clave
Citación
Centro
Facultades y escuelas::Facultad de Ciencias
Departamento
Física Fundamental