Machine learning-driven dielectric constant prediction for rock-salt structured microwave dielectric ceramics

Zhu Liu, Cheng Qian, Xiaolan Yang,
Machine learning-driven dielectric constant prediction for rock-salt structured microwave dielectric ceramics,
Journal of Alloys and Compounds,
Volume 1044,
2025,
184460,
ISSN 0925-8388,
https://doi.org/10.1016/j.jallcom.2025.184460.
(https://www.sciencedirect.com/science/article/pii/S0925838825060220)
Abstract: It is difficult to theoretically predict the dielectric constant for rock-salt type microwave dielectric ceramics owing to the multiple factors from the chemical compositions, crystal structure, and sintering conditions. Here, the Maximal Information Coefficient mutual information-based selection method is utilized to identify six key features associated closely with the crystal structure and microstructure of rock-salt microwave ceramic systems, and an eXtreme Gradient Boosting (XGBoost) model is adopted to achieve the optimal matching effect of dielectric constant, which is specially optimized by the INFO-based weighted vector means optimization algorithm. Results show that the sintering temperature, atomic ratio, and electronegativity are the most important factors influencing the prediction accuracy of the model, which contributes about 71 % to the prediction of εr value, where a small mean absolute error value of 0.5249 and a high coefficient of determination value of 0.9376 were achieved in the test dataset. This phenomenon indicates that the XGBoost model performs well using six features, which is further identified through experimental validation on two different material compositions, including LiMg8AlO10 and Li2SnO3 doped Li3MgNbO5 materials. Finally, a high efficiency and accuracy of the model for predicting the dielectric constant for rock-salt type microwave dielectric ceramics is developed, showing potential for discovering ceramic materials with a fixed dielectric constant range.
Keywords: Microwave ceramic; Rock-salt structure; Dielectric constant; Machine learning