IC-BTCN: A Deep Learning Model for Dropout Prediction of MOOCs Students

Abstract

Contribution: This study proposes a student dropout prediction model, named image convolutional and bi-directional temporal convolutional network (IC-BTCN), which makes dropout prediction for learners based on the learning clickstream data of students in massive open online courses (MOOCs) courses. Background: The MOOCs learning platform attracts hundreds of millions of users with in-depth teaching content and low-threshold learning methods. However, the high-dropout rate has always been its weakness compared with offline teaching. Intended Outcomes: The effectiveness of IC-BTCN model is evaluated on the KDD CUP 2015 dataset, including a large amount of clickstream data from the online learning platforms. The experimental results show that IC-BTCN model achieves an accuracy rate of 89.3%. Application Design: First, learning record data of students are converted into 3-D learning behavior matrix. Then, local features of the behavior matrix are extracted through convolutional techniques. These extracted learning features are then input into a temporal convolutional network to further refine the data. The temporal learning features of students are extracted through dilated causal convolution. Finally, a multilayer perceptron is used to derive the dropout prediction for students. Findings: Compared with three typical deep learning models, IC-BTCN model is advanced in accuracy and other evaluation indicators. On the premise of complying with the provisions of MOOCs platforms, the IC-BTCN model has good portability and practicability.