ABSTRACT
How early can we robustly identify the warning signs of poor academic performance and becoming ‘at-risk’ among university students? Using real-time behavioural data generated by a semester-long assessment and a machine learning approach, we find that online engagement alone can already predict whether a student will pass or fail a course with 85% precision within a few weeks into the beginning of term and can pinpoint the final course mark within a roughly 12% margin of error. Unsurprisingly, adding student background information translates into a sizeable 30% margin of error drop. Finally, we consistently see indicators related to regularity and depth of online engagement featuring prominently at the top of the academic performance predictor list, while among background elements only past academic ability seems to matter.
