Exploring Temporal Heterogeneous Graph Deep Learning and Machine Learning Models for Predicting Student Success

Predicting student success early and throughout the semester is critical to developing timely support strategies that help reduce dropout rates and encourage on-time graduation. While AI-powered systems show strong promise, key challenges remain, particularly in modeling the temporal evolution of student course performance and integrating the heterogeneous nature of educational data. This study uses temporal Heterogeneous Graph Neural Network (HGNN) models and temporal Machine Learning (ML) models that leverage temporal features, dynamic features, and metapath-based structures. To evaluate the impact of different model components, we conduct extensive experiments on the OULA dataset including simple vs. cumulative student performance features, ablation study on feature assignment, and metapath selection. The results indicate an improvement of up to 10.1% in F1 validation scores early in the semester when using temporal ML models compared to their static counterparts, and an improvement of up to 7.7% for temporal HGNN models over their static counterparts. These results demonstrate that temporal models effectively capture complex relationships within student data.