Prediction of at-risk students using learning analytics: A literature review

Abstract

Early identification of at-risk students has been recognised as being of the utmost importance for the provision of timely support. Despite the prevalent use of learning analytics in this regard, little attention has been paid to systematically surveying and summarising relevant latest work. To address the literature gap, this paper reviews the use of learning analytics to support prediction of at-risk students. The study covers 138 research articles published between 2013 and 2022 which were collected from the Scopus database. Through a content analysis approach, the relevant work was examined in terms of the prediction objectives, the types of data collected, the techniques used for prediction, and the metrics employed for evaluation of prediction performance. The findings reveal a strong scholarly interest in the prediction of students’ learning performance to identify those who are potentially at risk. The common types of data collected are related to students’ previous academic performance, socio-demographics, and learning behaviours, particularly those on learning management systems. The most frequently used techniques for prediction are decision trees, neural networks, and Bayesian networks. The results also show the widespread use of classification accuracy, recall, sensitivity, and true positive rate as the performance metrics. The findings contribute to advancing our understanding about the potential of learning analytics for at-risk student prediction, as well as informing the provision of timely and proper support for specific student groups.