ENGAGEMENT PROFILES AND ACADEMIC VULNERABILITY IN HIGHER EDUCATION: COUNTERFACTUAL RECOMMENDATIONS FOR STUDENT MONITORING

Abstract

This article analyzed how student engagement profiles, obtained through unsupervised clustering, can support the identification of academic vulnerability and the formulation of interpretable counterfactual recommendations for educational monitoring. The study adopted a quantitative, applied, and exploratory approach using secondary data from the Open University Learning Analytics Dataset (OULAD). The methodology involved preprocessing, selection of behavioral variables from the virtual learning environment, K-Means clustering, assessment of cluster quality and stability, and generation of counterfactual recommendations restricted to actionable variables. Demographic attributes and final performance were excluded from clustering and reserved for profile interpretation and external validation. The results revealed an engagement gradient, with low, medium, and high participation profiles. Low-engagement groups concentrated higher proportions of withdrawal and failure, indicating an association with academic vulnerability. The counterfactual recommendations made it possible to simulate transitions toward higher-participation profiles and proved feasible in most analyzed transitions. The study concludes that the proposed approach can support educational decision-making, provided its recommendations are understood as auditable statistical inputs rather than automatic prescriptions or causal inferences.