Abstract:

During their practice, educators often overlook the heterogeneity of possible learning strategies their students use when not in the classroom. However, a better understanding of this aspect can help identify the most effective learning strategies for a specific topic (or set of similar topics). Even more so, that understanding can inform course designers to create grading schemas that promote and reward the adoption of those learning strategies.

A key barrier lies in tracking student behaviour when not under direct supervision. Learning management systems (LMS) can help bridge this gap, as LMS logs record every student's interaction with the contents provided in it. These records can be transformed into timestamped sequences of clicks that reflect, albeit noisily and partially, the effort and learning strategies that students employ in their pursuit of academic success.

In this work, we used the Moodle logs generated by a sample of 3840 enrollments (made by 409 unique Bachelor's level students attending 57 unique courses) at a European information management school in 2020/2021. The first step was the conversion of the raw logs into a structured dataset whose rows represent one student's enrollment in one course. The second step was extracting and selecting variables representing three perspectives: Raw activity, Time on task, and Frequency. In the final stage, we combined all perspectives and used the K-Means algorithm to group similar enrollment types.

Our unsupervised model identified four distinct types of LMS usage strategies adopted by the students. The groups were then compared in terms of the average performance of the students following each of the strategies. These results can inform possible interventions that promote adopting adequate learning strategies and, ultimately, improve student performance.

Keywords:

Learning Management Systems, Clustering, Clickstream Data, Higher Education.