Grade prediction of weekly assignments in MOOCS: mining video-viewing behavior

David John Lemay, Tenzin Doleck

Abstract: Massive open online courses (MOOCs) hold the promise of democratizing the learning process. However, providing effective feedback has proven hard to offer at scale since most methods require a teacher or tutor. Leveraging big data in MOOCs offers a mechanism to develop predictive models that can inform computer-based pedagogical tutors. We review research on grade prediction and examine the predictive power of a model based on user video-watching behavior. In a MOOC organized around weekly assignments, we find that frequency of video viewing per week is a better predictor than individual viewing features such as plays, pauses, seeking, and rate changes. This finding is useful for MOOCs that use assignments for course evaluations in addition or to the exclusion of in-video quizzes for formative assessment. Engaging, well-crafted assignments in MOOCs have the potential of boosting student retention and course completion by fostering a deeper understanding through application and practice.

External IDs:dblp:journals/eait/LemayD20