Strategic use of predictive analytics for student retention in blended higher education: implications for education management

Abstract

This research investigates how predictive analytics can support student retention in blended higher education. Using the engagement and assessment data of 523 students from 3 universities, four machine learning models were developed. Of these four models, Multilayer Perceptron achieved the highest accuracy (70.0 per cent) and after handling class imbalance performed better at identifying students that are at risk of failure. Utilising theories of self-regulated learning, behavioural engagement, and academic integration, the study identifies the indicators that are most frequently associated with successful outcomes. Predictions of machine learning models were compared to the early predictions of instructors and revealed different error patterns, suggesting that models and instructors capture different dimensions of student engagement. Most significant predictors were learning management system activity, pacing, assessment behaviours, and attendance. Findings suggest that predictive analytics can support early warning systems as well as guiding institutional strategies and strengthening ethically responsible retention strategies in blended higher education.