Using Predictive Analytics to Support Students and Reduce Attrition: A Rapid Evidence Assessment

This paper advocates for a data-driven, proactive approach to identifying students at risk of disengagement and/or discontinuation. It aims to provide the evidence base for University of Westminster’s adoption of a predictive learner analytics model. Student disengagement is a complex, multi-faceted phenomenon influenced by multiple factors spanning academic performance, behavioural patterns, attitudinal factors, demographics, and institutional dynamics. Utilising a Rapid Evidence Assessment (REA) methodology, 56 peer-reviewed and industry studies and reports were analysed. This exercise identified 48 (dis)engagement indicators that should be considered when thinking about adopting a predictive model. Additionally, the REA highlighted numerous Machine Learning (ML) techniques used by Higher Education institutions to track student engagement and proactively ensure retention. These were discussed with Random Forest being highlighted as a precise and accurate technique that has been used by our UK HE peers. The paper also considered the importance of considering ethical and operational issues, from data architecture and governance to stakeholder buy-in, ethics, as well as data protection and privacy.