How Late is Too Late? Clustering Student Timing Behaviors to Predict Academic Success

This study examined temporal and behavioral engagement patterns in an asynchronous online learning environment using learning analytics. The sample consisted of 108 traditional undergraduate students enrolled in a course delivered through a Learning Management System (LMS). Three time-based indicators were derived from LMS logs: the time between release and first access, the time between first access and submission, and the time between submission and due date. To capture behavioral stability beyond average timing, additional variability-sensitive features were computed, including a regularity index, behavioral entropy, and deviation from the class mean. Cluster analysis identified three distinct engagement profiles, namely Strategic Delayers, Early Starters, and Last Minute Submitters, each associated with different academic outcomes. Regression analysis revealed that the duration between submission and due date and the duration between release and submission were the strongest predictors of course performance. Together, these variables explained 29% of the variance. The findings highlight that not all delayed engagement is detrimental and that temporal regularity and stability play critical roles in academic success. The study contributes to a behaviorally grounded understanding of engagement timing and offers implications for early warning systems, adaptive pacing, and formative feedback design in digital learning contexts.