Learning management systems generate trace data that researchers cluster to identify participation profiles. But a new preprint argues that unequal observation windows—where learners have different amounts of time in the course—can distort these profiles, making them reflect duration rather than actual participation.

To address this, the authors propose a heterogeneous consensus clustering approach that accounts for varying observation windows. They applied it to 16,448 complete learner-course records, though the abstract does not report specific outcomes.

The work highlights a methodological pitfall in educational data mining. If duration is not separated from engagement, interventions based on profile membership could target the wrong learners.