Research on the Prediction Model of Crystallinity of Natural Rubber Based on Gan-XGBoost Framework

The advent of big data era and prominence of artificial intelligence has catalyzed the applications of machine learning algorithms in various domains, including materials science, information technology, control engineering, and biomedicine. This paper explores the combination of machine learning and polymer material science by employing predictive models to ascertain the mechanical properties of natural rubber materials through computational simulations. The proposed solution is grounded in the GAN-XGBoost model framework, which leverages a data augmentation technique founded on Generative Adversarial Networks (GAN) to enhance the initial data set. Additionally, the XGBoost model, which utilizes treestructured algorithms, is implemented to expedite the prediction of crystallinity in these materials. To refine the insights gained, a weighted integration approach utilizing Shapley Additive Explanation (SHAP) analysis is adopted, enabling a nuanced understanding of how variations in phospholipid protein percentage, hydrogen bond strength, and non-hydrogen bond strength influence the crystallinity predictions of natural rubber materials under dynamic conditions. This model effectively integrates the crystallinity prediction algorithm for polymer materials through molecular dynamics simulations with the GAN-based data augmentation algorithm, thereby significantly enhancing the predictive accuracy of the XGBoost model.