Supplementary materials for "Strategic Use of Predictive Analytics for Student Retention in Blended Higher Education: Implications for Education Management"

This study examines how predictive analytics can support student retention in blended higher education. Using engagement and assessment data from 523 students at three universities, four machine learning models were developed to predict academic outcomes. Drawing on theories of academic integration, self-regulated learning, and behavioural engagement, the analysis identifies which engagement patterns are most closely linked to success. The Multilayer Perceptron achieved the most consistent performance (70.0% accuracy), with improved identification of at-risk students after addressing class imbalance. Compared with early instructor predictions, the model showed similar overall accuracy but revealed complementary error patterns, suggesting that human and machine judgments capture different aspects of student performance. Key predictors included learning management system activity, pacing, assessment behaviours, and attendance. Overall, the findings highlight the potential for predictive analytics to inform early-warning systems, guide institutional support, and strengthen ethically responsible retention strategies in blended learning environments.