Thermoelectric materials enable direct conversion of waste heat into electricity, but their rational design is hindered by the intrinsic coupling among the Seebeck coefficient (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">$\varvec{S}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">S</mi> </mrow> </math></EquationSource> </InlineEquation>), electrical conductivity (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">$\varvec{\sigma }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">σ</mi> </mrow> </math></EquationSource> </InlineEquation>), thermal conductivity (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">$\varvec{\kappa }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">κ</mi> </mrow> </math></EquationSource> </InlineEquation>), and figure of merit (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">$\varvec{ZT}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">ZT</mi> </mrow> </math></EquationSource> </InlineEquation>). Here, we develop a unified, composition-driven machine-learning framework for the simultaneous prediction of all four transport properties. A curated dataset of 4,251 samples, containing experimental data, was represented using 246 composition-based descriptors (215 elemental statistics and 31 physically informed proxies). Systematic benchmarking identifies ExtraTrees as optimal for <InlineEquation ID="IEq5"> <EquationSource Format="TEX">$\varvec{S}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">S</mi> </mrow> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq6"> <EquationSource Format="TEX">$\varvec{\sigma }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">σ</mi> </mrow> </math></EquationSource> </InlineEquation>, and <InlineEquation ID="IEq7"> <EquationSource Format="TEX">$\varvec{ZT}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">ZT</mi> </mrow> </math></EquationSource> </InlineEquation>, and XGBoost for <InlineEquation ID="IEq8"> <EquationSource Format="TEX">$\varvec{\kappa }$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold-italic">κ</mi> </mrow> </math></EquationSource> </InlineEquation>, achieving high predictive accuracy (<InlineEquation ID="IEq9"> <EquationSource Format="TEX">$\varvec{R^{2} = 0.953}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mi mathvariant="bold-italic">R</mi> <mn mathvariant="bold">2</mn> </msup> <mo mathvariant="bold">=</mo> <mn mathvariant="bold">0.953</mn> </mrow> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq10"> <EquationSource Format="TEX">$\varvec{0.918}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn mathvariant="bold">0.918</mn> </mrow> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq11"> <EquationSource Format="TEX">$\varvec{0.963}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn mathvariant="bold">0.963</mn> </mrow> </math></EquationSource> </InlineEquation>, and <InlineEquation ID="IEq12"> <EquationSource Format="TEX">$\varvec{0.927}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn mathvariant="bold">0.927</mn> </mrow> </math></EquationSource> </InlineEquation>) with minimal overfitting (<InlineEquation ID="IEq13"> <EquationSource Format="TEX">$\varvec{\Delta R^{2} < 0.08}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="bold">Δ</mi> <msup> <mi mathvariant="bold-italic">R</mi> <mn mathvariant="bold">2</mn> </msup> <mo mathvariant="bold"><</mo> <mn mathvariant="bold">0.08</mn> </mrow> </math></EquationSource> </InlineEquation>). SHAP-based interpretability reveals strong agreement with established thermoelectric physics, highlighting the roles of temperature, compositional complexity, electronic structure, and phonon-scattering descriptors in governing transport behaviour. High-throughput screening of 98,787 Materials Project compounds successfully recovers known high-performance thermoelectrics and identifies promising candidates for both near-room- and high-temperature applications. The proposed framework offers a physically interpretable and computationally efficient alternative to conventional first-principles approaches, enabling rapid and scalable discovery of advanced thermoelectric materials.

Ensemble machine-learning models for simultaneous prediction of thermoelectric transport properties

期刊论文
发表日期：

2026年7月1日

查看全文

Rumaisa Jan,
Sukanya Ghosh,
Seemin Rubab

摘要

Thermoelectric materials enable direct conversion of waste heat into electricity, but their rational design is hindered by the intrinsic coupling among the Seebeck coefficient ( $\varvec{S}$ $S$ ), electrical conductivity ( $\varvec{\sigma }$ $σ$ ), thermal conductivity ( $\varvec{\kappa }$ $κ$ ), and figure of merit ( $\varvec{ZT}$ $ZT$ ). Here, we develop a unified, composition-driven machine-learning framework for the simultaneous prediction of all four transport properties. A curated dataset of 4,251 samples, containing experimental data, was represented using 246 composition-based descriptors (215 elemental statistics and 31 physically informed proxies). Systematic benchmarking identifies ExtraTrees as optimal for $\varvec{S}$ $S$ , $\varvec{\sigma }$ $σ$ , and $\varvec{ZT}$ $ZT$ , and XGBoost for $\varvec{\kappa }$ $κ$ , achieving high predictive accuracy ( $\varvec{R^{2} = 0.953}$ $R^{2} = 0.953$ , $\varvec{0.918}$ $0.918$ , $\varvec{0.963}$ $0.963$ , and $\varvec{0.927}$ $0.927$ ) with minimal overfitting ( $\varvec{\Delta R^{2} < 0.08}$ $Δ R^{2} < 0.08$ ). SHAP-based interpretability reveals strong agreement with established thermoelectric physics, highlighting the roles of temperature, compositional complexity, electronic structure, and phonon-scattering descriptors in governing transport behaviour. High-throughput screening of 98,787 Materials Project compounds successfully recovers known high-performance thermoelectrics and identifies promising candidates for both near-room- and high-temperature applications. The proposed framework offers a physically interpretable and computationally efficient alternative to conventional first-principles approaches, enabling rapid and scalable discovery of advanced thermoelectric materials.