Modelling students’ behaviour: Mining and clustering digital learning paths to identify at-risk students

Abstract

The increasing adoption of digital learning environments has led to the generation of rich behavioural data from student interactions in Learning Management Systems. This study aims to identify distinct learning behaviour patterns and examine their association with course failure. Using Dynamic Time Warping, students are clustered based on their activity trajectories, followed by process mining to visualise differences in their learning pathways. The results reveal five learning strategy clusters, two showing a higher incidence of course failure. In the second phase, a Generalized Additive Model is applied to identify key behavioural characteristics of students who failed the course based on the derived process models. The findings highlight that execution-focused or low engagement patterns are more prevalent among unsuccessful students, while sustained, high individual commitment is associated with successful outcomes. Furthermore, the process models enable the identification of critical time periods and learning pathways characteristic of failure. By integrating trajectory-based clustering with a process-centric approach, the study provides a detailed characterization of behaviours associated with course failure and offers actionable insights for improving learning outcomes.