A Hybrid Deep Learning Model for MOOCs Dropout Prediction

Abstract

In recent years, Massive Open Online Courses (MOOCs) have attracted more and more learners due to its convenience and openness. However, the problem of high dropout rate has been difficult to solve, which hinders the further progress of MOOCs platforms. An accurate dropout prediction model that can effectively predict the dropout in MOOCs and intervene with students in advance is needed. Therefore, this paper proposes a novel hybrid model to predict learners' dropout behavior. Instead of manually extracting feature, this hybrid model designs a two-channel Convolutional Neural Network (CNN) to automatically extract useful feature from students' learning records, then employs Attention mechanism to obtain the important information. Finally, it applies Temporal Convolutional Network (TCN) to capture the relationships between hidden feature at different time scales. According to extensive experiments on the KDD CUP 2015 dataset, we can learn that the proposed model can achieve better results compared to other existing dropout prediction methods.