Modeling mathematics achievement with deep learning methods

Abstract

Deep learning methods are the subfield of the machine learning models that have spread rapidly in the field of engineering in the last decade. But, these methods are a fairly new in educational literature. The aim of this study was modeling and predicting mathematics achievement of successful and unsuccessful students via deep learning methods. For this purpose, Turkey's Programme for International Student Assessment (PISA 2018) survey data was used. Deep learning methods were displayed comparable performance to multi-layer perceptron and logistic regression. Jordan neural network method was found the most successful method among Elman neural network, Logistic regression and multi-layer perceptron methods with 0.826 accuracy and 0.739 area under curve scores. It was understood that deep learning methods can be used in the modelling and predicting of students' mathematics achievement.