Clustering Student Performance Across Changing Teaching Modalities: A Three-year Longitudinal Learning Analytics Study

Abstract

Learning Analytics research frequently examines learner behaviour and performance within a single cohort or teaching context, leaving limited evidence on whether comparable learner profiles recur across different teaching modalities. This study addresses this gap by analysing three consecutive cohorts (N=636) of the same university course delivered under fully online, blended, and on-site teaching conditions. Data included knowledge-domain mastery estimates and LMS activity logs used for clustering, while final course grades were used post hoc to characterise the resulting learner profiles. K-means clustering was used to identify and compare learner profiles, while non-parametric tests examined differences in mastery and final grades across cohorts. Three broadly comparable profiles emerged: higher-performing students (C1), failing students (C2), and struggling students comprising failing and low-passing students (C3). While their performance-based structure remained recognisable across cohorts, LMS activity patterns varied considerably. C2 consistently exhibited the lowest activity, whereas higher activity did not consistently distinguish higher-performing from struggling students. Significant differences in knowledge-domain mastery were observed across cohorts, with higher mastery levels in later cohorts in four of seven domains, while significant differences in final grades emerged primarily for the third cohort. The findings extend previous single-cohort research and demonstrate how learner profiles can inform differentiated educational support.

The authors' abstract, as published at the source. The Electronic Journal of e-Learning, 2026 · DOI ↗