<p>This study explores the application of deep learning, in particular graph neural networks (GNNs), to predict the composite performance of refractory and ceramic materials. The research aims to optimize material formulations for critical properties such as density, thermal stability, and mechanical strength. By leveraging a dataset compiled from both the literature and experimental measurements, an optimized GNN model was developed. This model dynamically updates molecular graph feature matrices through iterative layers, capturing complex interactions within material structures. The dataset was partitioned into three subsets: 80% for training, 10% for validation, and 10% for testing, ensuring a rigorous evaluation process. Key findings indicate that the optimized GNN model significantly surpasses traditional chemical analysis methods in prediction accuracy, evidenced by lower mean absolute error and higher coefficient of determination (<i>R</i>²) values. Data normalization techniques, including max-min scaling, were employed to enhance model robustness and generalization. These results demonstrate the feasibility and effectiveness of using deep learning for predicting material properties, thereby reducing empirical testing costs and time. Future work will focus on expanding the dataset to encompass a broader range of materials and refining the model to improve its generalization capabilities. This research paves the way for accelerated development and innovation in refractory and ceramic materials, with potential impacts on industries reliant on high-performance materials.</p>