MOOC Dropout Prediction Based on Multi-view Learning

Abstract

Massive Open Online Courses (MOOCs) provides a promising way to support education for all. Nonetheless, one central challenge is the remarkably high dropout rate, with completion rates for MOOC recently reported to be below 5%, and high dropout rates limiting their effectiveness. Building on the analysis of dropout as closely related to user and course interaction patterns, this paper presents an end-to-end multi-view interactive framework (EMIF) to predict user dropout in MOOC. We extract learning activity information, user enrolment information, and course knowledge information through a set of lD-CNNs, obtain mutual information of these and fuse the above information-view loss through a self-weighing method. Experimental evaluation of our solution on the KDDCUP 2015 and MOOCCube datasets showed better results (AUC & F1 scores). We have a better performance compared to several existing methods, without using different architectures for different datasets.