Student Behavior Evolution Prediction Via Temporal Graph Network Integration

With the growing availability of educational behavioral data, modeling the temporal evolution of student behavior has become increasingly critical for intelligent student management and early intervention. However, conventional methods often struggle to capture dynamic relational structures and identify behavioral inflection points. This paper proposes a student behavior evolution prediction model based on Temporal Graph Network (TGN) integration. By representing students as nodes and their time-stamped interactions as edges in a dynamic graph, the model enables continuous-time node embedding and trend-aware behavior forecasting. Experiments on real-world datasets from two universities in Guangdong Province demonstrate that the proposed approach outperforms baseline models in Accuracy, F1-Score, and AUC. The model exhibits superior sensitivity to behavioral transitions and demonstrates strong generalization performance. This study advances the application of graph neural networks in education and provides a robust foundation for developing early-warning systems and personalized student support strategies.