Broad Learning System for Predicting Student Dropout in Massive Open Online Courses

Abstract

At a time when the number of Massive Open Online Courses (MOOCs) users, courses and participating universities is increasing rapidly, a short-time training and reliable prediction model for MOOC extremely high dropout rates is needed. This paper proposes a MOOC dropout prediction model which is based on Broad Learning System (BLS) for MOOC dropout prediction. The model first maps the input into the feature node layer, then generates the enhanced node layer according to the feature node layer through activation, and finally performs linear transformation by combining the feature layer and the enhancement layer. The output layer is used for dropout prediction. Experiments are carried out on the dataset provided by KDD CUP 2015, and the experimental results show that the BLS significantly reduces training time and has a better prediction of dropouts than other mainstream research methods.