New recommender system for enhancing predictions’ performances in e-learning systems using learning analytics indicators

Abstract

With recent advances in machine learning, the teacher can easily predict students at risk of failure, their performance, grades, and many other tasks on an e-learning system. However, these predictions do not always result in the best possible performance. In some cases, many errors are generated, and the teacher cannot take corrective actions in time. For this reason, we propose in this paper a new recommender system that focuses on the optimization of prediction performance using Learning Analytics indicators. This system proposes to the user the best set of learning indicators producing the best possible prediction performance. Indeed, the proposed recommendation system has been tested for many prediction tasks and can be adapted to many machine learning algorithms. The results show that the system is suitable for several prediction tasks as well as classification and regression algorithms. Thus, the recommender was evaluated using many metrics such as accuracy, ROC-AUC score, R2 score, and RMSE, among others, resulting in 99% and 98% accuracy for binary classification and multi-class classification tasks, respectively. As well as an R2 score of 92% for the regression task.