Mining Students' Digital Behaviors in Class to Create an Earlier Warning System of Student Success

In this paper, we present how students' digital behaviors can be used to describe and predict student success in an introductory atmospheric science course. The results include two major findings. First, we found behavioral signals like the number of correct responses to in-class questions, the number of confusing slides, and the number of viewed slides and videos are stable predictors of student success across different periods of a semester. Second, using both behavioral signals and incoming profiles, our weekly forecast model on student success achieved a 72% prediction accuracy by the end of the third week of classes and before the first exam. We believe these findings can set the stage for development of "earlier" warning systems that can identify students at risk before their first major assessment and speculates about how to use this forecast to motivate changes in student behaviors.