Using Meta-Learning to predict student performance in virtual learning environments

Abstract

Educational Data Science has meant an important advancement in the understanding and improvemen of learning models in recent years. One of the most relevant research topics is student performance prediction through click-stream activity in virtual learning environments, which provide abundant information about their behaviour during the course. This work explores the potential of Deep Learning and Meta-Learning in this field, which has thus far been explored very little, so that it can serve as a basis for future studies. We implemented a predictive model which is able to automatically optimise the architecture and hyperparameters of a deep neural network, taking as a use case an educational dataset that contains information from more than 500 students from an online university master’s degree. The results show that the performance of the autonomous model was similar to the traditionally designed one, which offers significant benefits in terms of efficiency and scalability. This also opens up interesting areas of research related to Meta-Learning applied to educational Big Data.