Massive Open Online Courses (MOOCs) are one of the fastest growing and most popular phenomena in e-learning. Universities around the world continue to invest to create and maintain these online courses. Reuse of material from previous courses is a shared practice that helps to reduce production costs and enhance future offerings. However, such re-runs still experience a high number of users not completing the courses, one of the most compelling issues of MOOCs. Hence, this research utilizes the information from the first run of a MOOC to predict the behavior of the users on a successive offering of the same course. Such information allows instructors to identify users at risk of not to finishing and helps to improve successive offerings. To this end, we analyze two successive offerings of the same MOOC, created by Curtin University on the edX platform. We extract features from the original run of the MOOC and predict dropouts on its re-run. We experiment with a Boosted Decision Tree and consider two different approaches: a varying percentage of users active time and users' first week of interactions with the MOOC. We obtain an accuracy of 0.8 when considering 10% of users active time or the first five days after users initial interaction. We also identify a set of features that are likely to indicate whether users will attrite in the future. Moreover, we discover typical patterns of interactions and notice a first set of tools that account for most interactions and a second one that is practically overlooked by users. Finally, we discover subgroups among the Dropouts characterized by similar behaviors. Such knowledge can be used to shape the structure of courses accordingly.
Fields of Expertise
- Information, Communication & Computing