Physics-inspired feature engineering and interpretable machine learning for thermoelectric properties of multicomponent materials
4 affiliations
- 1Faculty of Mechanical, Electrical and Computer Engineering, Van Lang School of Technology, Van Lang University, Ho Chi Minh City, 70000, Viet Nam
- 2Laboratory for Computational Physics, Institute for Computational Science and Artificial Intelligence, Van Lang University, Ho Chi Minh City, 70000, Viet Nam
- 3Department of Materials and Nanotechnology, SINTEF Industry, Trondheim, NO-7034, Norway
- 4Porelab, Department of Chemistry and Biomedical Science, Norwegian University of Science and Technology, NTNU, Trondheim, NO-7491, Norway
Abstract
Machine learning can accelerate thermoelectric materials discovery, yet most models remain “black boxes” lacking physical insight. We present an interpretable framework coupling Extra Trees regression with SHAP analysis to predict electrical conductivity (EC), Seebeck coefficient (SC), and thermal conductivity (TC) across 3879 multicomponent compositions. Physics-inspired feature engineering expands descriptors to 73–84 per target via polynomial lattice-distortion terms, cross-interactions, and compositional disorder statistics. Random-split test-set values reach 0.961, 0.965, and 0.958 for EC, SC, and TC, respectively, surpassing a lightweight deep neural network. Composition-wise splitting reduces to 0.822, 0.871, and 0.876, indicating limited generalisation to unseen compositions. SHAP analysis reveals property-specific associations: elemental descriptors dominate all targets, lattice-distortion terms influence EC, and heavy-element features affect TC. As a consistency check, a ZT-proxy computed from the independently predicted transport properties agrees reasonably well with experimental values on the test set ( = 0.84); it is intended as a screening aid rather than a verified figure of merit. This interpretable framework supports screening and prioritisation backed by, not replacing, experimental validation.
Related
- Nano-Based Advanced Thermoelectric Design: Materials Engineering and Thermal Management Applications
- Synergistic Grain Boundary and Phonon Engineering Enables Ultrahigh Thermoelectric Performance in Mg 3 (Sb, Bi) 2 ‐Based Materials
- Enhancing Near‐Room‐Temperature Thermoelectric Performance of n‐Type Mg3(Sb, Bi)2‐Based Materials via ZrB2 Heterointerface Engineering
- Supercapacitor Materials: Structure, Properties, and Applications for Energy Storage in Engineering Systems
- Upcycling Recycled PET into High-Performance Polyurethane Coating and Engineering Materials: Effect of NCO Index on Mechanical, Stability, and Surface Properties
BibTeX
@article{ho2026physicsinspired,
title = {{Physics-inspired feature engineering and interpretable machine learning for thermoelectric properties of multicomponent materials}},
author = {Thi Huynh Ho and Nguyen-Hieu Hoang and Thuat Thanh Trinh},
journal = {Materials \& Design},
year = {2026},
month = jun,
doi = {10.1016/j.matdes.2026.116533},
url = {https://doi.org/10.1016/j.matdes.2026.116533},
note = {CC BY},
}
RIS
TY - JOUR
TI - Physics-inspired feature engineering and interpretable machine learning for thermoelectric properties of multicomponent materials
AU - Thi Huynh Ho
AU - Nguyen-Hieu Hoang
AU - Thuat Thanh Trinh
PY - 2026
DA - 2026/06/30
JO - Materials & Design
DO - 10.1016/j.matdes.2026.116533
UR - https://doi.org/10.1016/j.matdes.2026.116533
AB - Machine learning can accelerate thermoelectric materials discovery, yet most models remain “black boxes” lacking physical insight. We present an interpretable framework coupling Extra Trees regression with SHAP analysis to predict electrical conductivity (EC), Seebeck coefficient (SC), and thermal conductivity (TC) across 3879 multicomponent compositions. Physics-inspired feature engineering expands descriptors to 73–84 per target via polynomial lattice-distortion terms, cross-interactions, and compositional disorder statistics. Random-split test-set values reach 0.961, 0.965, and 0.958 for EC, SC, and TC, respectively, surpassing a lightweight deep neural network. Composition-wise splitting reduces to 0.822, 0.871, and 0.876, indicating limited generalisation to unseen compositions. SHAP analysis reveals property-specific associations: elemental descriptors dominate all targets, lattice-distortion terms influence EC, and heavy-element features affect TC. As a consistency check, a ZT-proxy computed from the independently predicted transport properties agrees reasonably well with experimental values on the test set ( = 0.84); it is intended as a screening aid rather than a verified figure of merit. This interpretable framework supports screening and prioritisation backed by, not replacing, experimental validation.
SN - 0264-1275
SN - 1873-4197
ER -