Supporting Group Formation in Ongoing MOOCs using Actionable Predictive Models

Abstract

Although the massive and open nature of MOOCs necessitates more technology support for instructors, existing prediction research has been barely capable of offering real-world solutions. One critical case where MOOC instructors could benefit from real-time technological support is the design of collaborative activities, in particular group formation. In this regard, this research work investigated the use of in-situ learning technique to produce useful and actionable information that could assist instructors in group formation while the course continues. Focusing on a particular MOOC context, a predictive model was created to compute the probability that students would participate in group discussions or not. Using these probability scores, actual group behavior was also predicted for three cases: at least 2, 3 or 4 different students would post in group discussions. According to the results, the model was able to accurately predict individual student behavior as well as group behavior before the actual collaborative activity had taken place, suggesting its potential for real-time use. Future research involves the exploration of other approaches for creating actionable predictions and the application of the predictions in practice in an ongoing MOOC.