MOOC Student Success Prediction Using Knowledge Distillation

Massive Open Online Courses (MOOCs) have boomed in recent years because learners can arrange learning at their own pace. Due to the low success rate of MOOCs, student success prediction has received much attention recently. Yet many utilize post-hoc prediction architectures, where model fitting requires features which are not knowable until a course completes. These works do not match the actual environment in which the prediction model is used, which may lead to unsatisfactory results. In order to build a reliable model that can be used in the actual environment, we propose a new student success prediction algorithm based on knowledge distillation. This algorithm can make the model obtain good prediction results with only a few basic features, and solve the problem of using features which are not knowable until a course completes.