Implementación de algoritmos predictivos con Deep Learning a través del uso de la librería Keras

Abstract

Abstract:Due to the growth in popularity of Deep Learning in the scientific environment in these years within the field of predictive algorithms and in view of the need to know how to improve the experience of students who attend a MOOC, this end-of-degree work seeks to determine if the Keras bookstore can mitigate this need. To achieve this, a Dataset is used with many logs that indicate the different activities that a student develops in the Open EdX portal of the University. Different models of neural networks were developed with multiple configurations that, after adjusting, allow to predict with a high degree of accuracy. The neural networks use CSV files that are understood by the machine sending back a prediction that can be directed to the operator in the form of CSV. The results obtained show that Deep Learning is a great option when trying to predict results in an academic environment. For a better management of the project, a three-layer architecture and a methodology derived from the CRISP-DM methodology are used.