Thermoelectric Properties in Skutterudite Materials: Integrating Experimental Data, Density Functional Theory, and Machine Learning

An XGBoost regression model has been utilized to accurately predict the temperature-dependent thermal conductivity (κ), electrical conductivity (σ), Seebeck coefficient (S), and figure of merit (ZT) of skutterudite materials from 50 to 1000 K. Shapley values, derived from cooperative game theory, were employed to quantify the contributions of various material descriptors, providing valuable insights into the underlying correlations and trade-offs between different features and thermoelectric properties. The analysis revealed crucial factors influencing the thermoelectric performance of skutterudites, including temperature, compositional attributes, electronic configurations, and structural properties. The model predicts a room-temperature ZT of approximately 0.45 for Nd(CoP₃)₄, a promising thermoelectric material that has not yet been experimentally investigated. In addition, the prediction of half-metallic ferromagnetic behavior in Nd(CoP₃)₄ by first-principles density functional theory (DFT) with Hubbard corrections has led to the investigation of its thermoelectric properties. At elevated temperatures, a strong correlation was observed between XGBoost predictions and DFT calculations for the σ and S of Nd(CoP₃)₄. The agreement between predicted and calculated values was less pronounced, though still reasonable, for κ and ZT. By integrating machine learning with fundamental materials science principles, this study paves the way for accelerated discovery and optimization of high-performance thermoelectric materials, contributing to the advancement of sustainable energy technologies and the pursuit of a more energy-efficient future.