Research on MOOC dropout prediction by combining CNN-BiGRU and GCN

To tackle the widespread issue of high dropout rates on MOOC platforms, this study presents a dropout prediction approach leveraging Graph Convolutional Networks (GCN). By analyzing learners' behavioral patterns at multiple temporal scales, the proposed method aims to identify potential dropout tendencies in a timely manner, enabling the adoption of targeted preventive or intervention strategies. Specifically, the proposed method automatically extracts continuous features from learners' activity logs over a specified period, utilizing these as independent variables for constructing a MOOC dropout prediction model. Initially, the behavioral data collected by the learning platform is transformed into time series data. A CNN-BiGRU network is then employed to extract features, generating feature vectors that effectively capture the temporal dynamics within the learners' activity patterns. These multi-dimensional feature vectors are subsequently used as node features in a Graph Convolutional Network (GCN), where two GCN layers establish intrinsic relationships between the features while preserving essential information through vectorization. This model allows MOOC instructors and course designers to monitor learners' progress in real time across different time intervals, enabling dynamic tracking of dropout behavior at various stages of the learning process. Experimental results on a large-scale MOOC public dataset demonstrate that the proposed method achieves superior dropout prediction accuracy compared to other state-of-the-art approaches.