Predicting Certification in MOOCs based on Students’ Weekly Activities

Abstract

Massive Open Online Courses (MOOCs) have been growing rapidly, offering low-cost knowledge for both learners and content providers. However, currently there is a very low level of course purchasing (less than 1% of the total number of enrolled students on a given online course opt to purchase its certificate). This can impact seriously the business model of MOOCs. Nevertheless, MOOC research on learners’ purchasing behaviour on MOOCs remains limited. Thus, the umbrella question that this work tackles is if learner’s data can predict their purchasing decision (certification). Our fine-grained analysis attempts to uncover the latent correlation between learner activities and their decision to purchase. We used a relatively large dataset of 5 courses of 23 runs obtained from the less studied MOOC platform of FutureLearn to: (1) statistically compare the activities of non-paying learners with course purchasers, (2) predict course certification using different classifiers, optimising for this naturally strongly imbalanced dataset. Our results show that learner activities are good predictors of course purchasibility; still, the main challenge was that of early prediction. Using only student number of step accesses, attempts, correct and wrong answers, our model achieve promising accuracies, ranging between 0.81 and 0.95 across the five courses. The outcomes of this study are expected to help design future courses and predict the profitability of future runs; it may also help determine what personalisation features could be provided to increase MOOC revenue