The data of "Artificial intelligence-based screening of dielectric properties in binary ionic‑doped yttrium iron garnets"

It primarily consists of the training data, descriptor correlation analysis results, cross-validation results, and model hyperparameter optimization results.The training data use the dielectric constants obtained from first-principles calculations as the target variable, together with six physically meaningful feature descriptors constructed from the compositions of the doped systems, including first ionization energy, electronegativity, polarizability, ionic radius, coordination number, and magnetic moment.

Among these descriptors, first ionization energy, electronegativity, polarizability, and ionic radius are characterized by the differences in the corresponding properties between the dopant and substituted elements, while coordination number and magnetic moment are used to describe the local geometric environment and magnetic characteristics induced by doping. Each data entry corresponds to a specific doped YIG composition and contains its six descriptors and the corresponding DFT-calculated dielectric constant.During model construction and feature selection, the dataset also includes descriptor correlation analysis based on the Pearson correlation coefficient (PCC), which is used to identify correlations among the descriptors and reduce the influence of redundant features on model prediction performance.