Multi-Graph Spatial-Temporal Synchronous Network for Student Performance Prediction

Abstract

In the realm of intelligent education, which is crucial for fostering sustainable student growth, predicting student performance stands out as a pivotal element. At its heart, the challenge of forecasting academic success lies in unraveling the complex, hidden relationships within performance data. While numerous investigations have addressed this challenge, existing approaches often overlook comprehensive, multi-perspective modeling and fail to capture the intricate spatial and temporal dependencies synchronously. To bridge these gaps, this study introduces a multi-graph spatial-temporal synchronous network (MGSTSN) to enhance the precision of performance predictions. By innovatively crafting spatial trend and pattern graphs alongside temporal causal graphs, this approach enables a holistic representation of the diverse spatial-temporal dynamics at play. Furthermore, the development of dual spatial-temporal synchronous graphs and their allied synchronous modules marks a novel strategy for simultaneously learning the interplay between spatial and temporal factors affecting student performance. Rigorous evaluations on authentic datasets reveal that MGSTSN significantly outperforms existing models, demonstrating an 8.17% to 11.34% enhancement across various metrics. This validates MGSTSN’s advanced capability in capturing the multifaceted nature of student performance data.

Keywords

• Graph convolutional networks
• multiple perspectives
• student performance
• prediction model