Predicting the academic performance of international students on an ongoing basis

Abstract

The academic success of international students is crucial for many tertiary institutions. Early predictions of students’ learning outcomes allow for targeted support and therefore improved success rates. In this study, international students’ demographic information, past academic histories, weekly class attendance records, and assessment results in an ongoing course were used to develop models to predict student success and failure in the course on a weekly basis. The prediction models were produced with three decision tree classification algorithms: REPTree, J48 tree, and LMT on the data-mining platform WEKA. Of these, the LMT algorithm has the highest level of accuracy, but the REPTree and J48 models are simpler and easier to interpret. While the accuracies of all three models are above 75%, further research is needed to more accurately predict student failure at early stages.