OU Analyse — The Open University's Predictive Learning-Analytics Early-Warning System

OU Analyse is a machine-learning early-warning system built by the Open University's (UK) Knowledge Media Institute that produces weekly predictions of which distance-learning students are at risk of failing an upcoming assessment, based on demographic data and clickstream interactions with the university's Virtual Learning Environment. Predictions are shared with course tutors and Student Support Teams to prioritise outreach.

The system aims to enable timely, targeted support for students identified as at risk, with the university describing timely intervention as the goal the predictions enable rather than a separately quantified causal outcome.

The system has run in continuous production since spring 2014, when it was piloted on two introductory courses (roughly 1,500 and 3,000 students); it expanded to 18 courses by autumn 2014/spring 2015 and has continued operating institution-wide for over a decade.

The underlying methodology and dataset (OULAD) were published in Nature's peer-reviewed journal Scientific Data in 2017, covering 22 courses, 32,593 students and over 10.6 million logged VLE-interaction records - one of the largest openly published learning-analytics datasets, widely reused by other researchers. Subsequent peer-reviewed research has examined prediction-accuracy trade-offs (precision rising but recall falling as more within-course data accumulates), how tutors use and trust the dashboards, and algorithmic fairness across student demographic groups.

The source material is explicit that what has been demonstrated is prediction accuracy and sustained operational use across many courses and years, not a randomised or quasi-experimental causal estimate of how much the predictions themselves raise pass or retention rates relative to a comparison group.