Machine learning assisted τf value prediction of ABO3-type microwave dielectric ceramics

In this work, we proposed a machine learning framework to predict the temperature coefficient of resonant frequency (τ f or TCF) values of ABO 4 -type ceramics. Leveraging 125 experimentally reported materials, a rigorous feature screening strategy integrating redundancy reduction with multiple feature selection methods was employed to identify a physically meaningful descriptor set. Molecular dielectric polarizability (pm), bond valence sum (bvs), degree of covalence (cvd), and space group (sp) were identified as dominant factors governing the τ f value. Notably, pm plays the most crucial role through the ion polarizability dilution theory. The Extreme Gradient Boosting (XGB) model exhibited the best predictive performance on the testing set (R 2 = 0.8539, RMSE = 6.68×10 –6 °C –1 ). The model was further applied to the Materials Project database for high-throughput screening, revealing quantitative descriptor windows associated with near-zero τ f . These insights provide practical guidance for the rational design of stable microwave dielectric ceramics .