Adopting Automatic Machine Learning for Temporal Prediction of Paid Certification in MOOCs

AbstractMassive Open Online Course (MOOC) platforms have been growing exponentially, offering worldwide low-cost educational content. Recent literature on MOOC learner analytics has been carried out around predicting either students’ dropout, academic performance or students’ characteristics and demographics. However, predicting MOOCs certification is significantly underrepresented in literature, despite the 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) and its financial implications for providers. Additionally, the current predictive models choose conventional learning algorithms, randomly, failing to finetune them to enhance their accuracy. Thus, this paper proposes, for the first time, deploying automated machine learning (AutoML) for predicting the paid certification in MOOCs. Moreover, it uses a temporal approach, with prediction based on first-week data only, and, separately, on the first half of the course activities. Using 23 runs from 5 courses on FutureLearn, our results show that the AutoML technique achieves promising results. We conclude that the dynamicity of AutoML in terms of automatically finetuning the hyperparameters allows to identify the best classifiers and parameters for paid certification in MOOC prediction.KeywordsMOOCsCertification predictionAutoMLAuto-sklearn