XGBoost-Based Modeling of Electrocaloric Property: A Bayesian Optimization in BCZT Electroceramics

Electrocaloric materials, which exhibit adiabatic temperature change under an applied electric field, are promising for solid-state cooling technologies. In this study, the electrocaloric response of lead-free Ba_xCa_1-xZr_yTi_1-yO₃ (BCZT) ceramics was modeled to investigate the effects of composition, processing, and measurement conditions on performance. A high-accuracy XGBoost regression model (R² = 0.99, MAE = 0.02 °C) was developed using a dataset of 2188 literature-derived data points to predict and design the electrocaloric response of BCZT ceramics. The feature space incorporated compositional ratios, processing parameters, measurement settings, and atomic-level Magpie descriptors, along with Curie temperature to account for phase-transition behavior. Feature importance analysis revealed that electric field, measurement temperature, and proximity to the Curie point are the most critical factors influencing ΔT_EC. Bayesian optimization was applied to navigate the design space and identify performance maxima under unconstrained and realistic constraints, offering valuable insights into the nonlinear interactions governing electrocaloric performance. Under room temperature and moderate-field conditions (24 °C, 40 kV/cm), the optimized ΔT_EC achieved a value of 1.03 °C for Ba_0.85Ca_0.15Zr_0.40Ti_0.60, to be processed at 1090 °C for 3 h during calcination, 1300 °C for 2 h during sintering. By integrating experimental insight with machine learning and optimization, this study offers a refined, interpretable framework for accelerating the design of high-performance electrocaloric ceramics while reducing the experimental workload.