Using Feature Interaction for Mining Learners' Hidden Information in MOOC Dropout Prediction

Massive open online courses (MOOC) are increasingly prevalent as a result of the rise in internet usage in recent years. However, the current development of MOOC is being severely hampered by the high dropout rates. The primary research goal of this work is to develop prediction models to identify students who are likely to exhibit dropout behavior in advance. In this paper, we propose the Cross-TabNet, which efficiently learns feature-hidden information by explicit feature interaction and uses sequential attention-based TabNet for classification. The experimental results demonstrate that it outperforms existing machine learning and deep learning methods.