A Nonlinear State Space Model for Identifying At-Risk Students in Open Online Courses

How to identify at-risk students in open online courses has received increasing attention, since the dropout rate is unexpectedly high. Most prior studies have focused on using machine learning techniques to predict student dropout based on features extracted from students’ learning activity logs. However, little work has viewed the dropout prediction problem as a sequence classification problem in the consideration that the dropout probability of a student at the current time step can be likely dependent on her/his engagement at the previous time step. Therefore, in this paper, we propose a nonlinear state space model to solve this problem. We show how students’ latent states at different time steps can be learned via this model, and demonstrate its outperforming prediction accuracy relative to related methods through experiment.