Deep Learning to Predict At-Risk Students’ Achievement in a Preparatory-year English Courses

Abstract

Predicting learners’ final course achievement is most of the time based on the grades they get on their graded course activities. Thus, it is of great importance for both students and higher education institutions to detect risk instances which can be addressed by the academic institution to support students’ success and academic advancement. In this context, Learning Analytics (LA), which represents learners’ behavior inside Learning Management Systems (LMS), and Deep Learning (DL) techniques come into play as academic data, which can be used to predict learners’ future achievements. It is not surprising that at-risk profiling becomes necessary when there are large numbers of students taking a preparatory course online, for example, where instructors fail to monitor their progress in real-time. Thus, the proposed study aims to utilize neural networks (vRNN, LSTM, and GRU); to build models that predict students’ final grade by classifing them as pass or fail based on their assessment grades. In the training process, the three models, alongside a baseline Multilayer Perceptron (MLP) classifier, were trained on four datasets illustrating students’ LMS activity and final grade results in a two-module English preparatory course in King Abdulaziz University (KAU). The datasets were collected during the first and second semesters of 2021. Results indicate that though all of the three DL models performed better than the MLP baseline, the GRU model achieved the highest classification accuracy on three datasets: (ELIA 103-1, 103-2, and 104-1) with the accuracy values of 92.21%, 97.75%, and 94.34%, respectively. On ELIA 104-2 dataset, both vRNN and LSTM achieved 99.89% accuracy. Considering the prediction of at-risk students, the three DL models achieved high recall values ranging from 65.38% to 99.79. %