Discovering ABO3-type perovskite with different dielectric constants via intelligent optimization algorithm

Nowadays, dielectric materials are playing an increasingly important role in various fields. A high dielectric constant (D) can store more charge per unit volume, improving performance, reducing device size, lowering D limit cross communication, and enabling better packaging of devices. Differentiating high D and low D has been recognized as a significant concern in electronics. However, calculating the dielectric constant from first principles is notoriously difficult. Therefore, it is essential to find important descriptors for predicting the dielectric constant (D) of different dielectric materials. In this work, a novel intelligence optimization approach was proposed based on data-driven methods to predict the dielectric constant (D) of ABO3-type perovskites. By applying the machine learning (ML) algorithm, key features strongly correlated with D were identified. To reduce feature dimension, Random Forest Regression-Gradient Boosting Regressor (RFR-GBR) feature screening, sure independence screening, and the sparsifying operator approach were employed to compress the feature set for creating valid descriptors. Furthermore, the Shapley additive explanation technique was used to reveal the scaling relation between the dielectric constant and the identified descriptors for predicting the D of ABO3-type perovskites. In addition, a hybrid artificial rabbits optimization algorithm and random forest regression were developed for predicting D, achieving remarkable predictive performance with an R2 score of 0.95, MAE of 0.23, and RMSE of 0.108 using five-fold cross-validation. Ultimately, from a pool of 300 candidate materials, we screened and identified two potential dielectric perovskites with different D values. The proposed framework will facilitate D prediction technology for the discovery of dielectric perovskite materials with favorable performance.