Popularity prediction in MOOCs: a case study on Udemy

Massive Open Online Courses (MOOCs) have dramatically changed how people access education. Though substantial research works have been carried out to improve students’ learning experiences, very little attention was directed to the characterization and identification of quality MOOCs for students to undertake (e.g., those with a large enrolment of students), which, we argue, is vital to empower students to make use of MOOCs to reskill and upskill. To fill the gap, this study aimed to investigate the extent to which ML models can be used to automatically identify the popularity of a MOOC before or upon its publication. Specifically, we collected data about more than 50K courses from Udemy, based on which we engineered a total of 21 features as input to four widely-used ML models for MOOC popularity prediction, namely Linear Regression, Random Forests, XGBoost, and Multi-Layer Perceptron Neural Network. Through extensive evaluations, we demonstrated that (i) XGBoost gave the best performance in predicting MOOC popularity; (ii) features like the number of captions and enrolment fee were strongly correlated with MOOC popularity; (iii) the prediction results were mostly inferior to those reported on predicting the popularity of social media posts and news articles, and thus more research effort is needed to boost the prediction performance.