A survey on predicting at-risk students through learning analytics

This paper analyses the adoption of learning analytics to predict at-risk students. A total of 233 research articles between 2004 and 2023 were collected from Scopus for this study. They were analysed in terms of the relevant types and sources of data, targets of prediction, learning analytics methods, and performance metrics. The results show that data related to students' academic performance, socio-demographics, and learning behaviours have been commonly collected. Most studies have addressed the identification of students who have a higher chance of poor academic performance or dropping out of their courses. Decision trees, random forests, and artificial neural networks are the most frequently used techniques for prediction, with ensemble methods gaining popularity in recent years. Classification accuracy, recall, sensitivity, and true positive rate are commonly used as performance metrics for evaluation. The results reveal the potential of learning analytics for informing timely and evidence-based support for at-risk students.