Research published in the International Journal of Innovation and Learning has reviewed the way we identify “at-risk” students. The study offers new insights into how educational support systems might be improved. Kam Cheong Li, Billy Tak-Ming Wong, and Maggie Liu of Hong Kong Metropolitan University in Homantin, Hong Kong, China, carried out an extensive review of some 233 papers from 2004 to 2023. They looked for common data types, sources, prediction targets, learning analytics methods, and performance metrics and how they are used in this field. The aim being to find the optimal approaches to identifying at-risk students effectively and efficiently.
Cheong and colleagues discuss how learning analytics involves the collection, analysis, and interpretation of data from various sources such as student performance, behaviour, and engagement. If it is possible to more effectively identify students at risk of academic failure, dropout, or other issues, then timely interventions might be made to address the problems those students are facing and help them make their way through academia successfully or guide them towards alternatives if appropriate.
The analysis highlights that data related to students’ academic performance, socio-demographic factors, and learning behaviour have traditionally been the mainstay of at-risk predictions. Most studies, the team reports, focus on identifying students at risk of poor academic performance or dropping out using decision trees, random forests, and artificial neural networks. They add that ensemble methods have come to the fore recently. The team has evaluated these various techniques using performance metrics such as classification accuracy, recall, sensitivity, and true positive rate.
This paper addresses the limitations of earlier literature reviews in this area by providing a comprehensive survey that includes a broad range of data sources and analytics techniques. The findings suggest that using multiple data types and combining various analytics methods and metrics can enhance the accuracy of at-risk student predictions. This emphasizes the need for a multifaceted approach to effectively identify and support such students.
Li, K.C., Wong, B.T-M. and Liu, M. (2024) ‘A survey on predicting at-risk students through learning analytics’, Int. J. Innovation and Learning, Vol. 36, No. 5, pp.1–15.