Application Research of Bayesian Optimization-based Graph Convolutional Neural Networks in Academic Warning Systems

In higher education, increasing attention is being given to “struggling students” as a distinct group. Current academic warning systems, which react after issues arise, hinder timely and effective interventions, highlighting the need for a proactive and precise warning system. This study introduces a Bayesian Optimization-based Graph Convolutional Neural Network (GCN) model to enhance academic warning systems' performance. Bayesian optimization is used to fine-tune GCN hyperparameters, automatically identifying the optimal model configuration to improve accuracy and robustness. By analyzing students' learning logs and creating interaction graphs between students and knowledge points, the model extracts embeddings through graph convolution. A Bayesian optimization strategy iteratively searches for optimal neural network structures and parameters like learning rate. Experiments with a dataset of 5,000 students and 1,000 knowledge points show that incorporating Bayesian optimization significantly improves key metrics such as recommendation accuracy and recall rate, and offers enhanced robustness in sparse data and cold start scenarios, providing a novel optimization approach for academic warning systems in complex networks.