Fig. 1: Schematic illustrating the exploring of high energy-storage performance BNT-based high-entropy compositions through a combination of machine learning and experimental verification.

a ABO₃ perovskite structure combined with the elements during machine learning. b Representative descriptors (c) Pearson correlation matrix for 60 descriptors. d Model selection through cross-validation R² score including training and testing in different models. e Performance of using Random Forest (RF) model on the training data using the 11 descriptor sets. f Result of descriptors importance. g Comparison of important descriptors corresponding to different A-site and B-site elements. h Candidate compositions with the highest W_rec obtained from element and proportion filtering. i XRD patterns, j P–E loops measured at low E, and k bandgap and P_m of A, B, C, and D samples.