Robust Predictive Models on MOOCs : Transferring Knowledge across Courses

As MOOCs become a major player in modern education, questions about how to improve their effectiveness and reach are of increasing importance. If machine learning and predictive analytics techniques promise to help teachers and MOOC providers customize the learning experience for students, differences between platforms, courses and iterations pose specific challenges. In this paper, we develop a framework to define classification problems across courses, provide proof that ensembling methods allow for the development of high-performing predictive models, and show that these techniques can be used across platforms, as well as across courses. We thus build a universal framework to deploy predictive models on MOOCs and demonstrate our case on the dropout prediction problem.