Machine-learning prediction of thermoelectric properties using high-entropy descriptors

This study integrates high-entropy (HE)–based descriptors with tree-based machine-learning models to predict electrical conductivity, Seebeck coefficient, and thermal conductivity in multicomponent thermoelectric materials. A curated dataset of 880 compositions was used to train Decision Tree, Random Forest, and Extreme Gradient Boosting (XGB) models across multiple train–test splits. XGB consistently achieved the best performance, with R 2 test values exceeding 0.90 for all target properties. Feature-importance analysis identifies atomic-size mismatch (δ) as the dominant descriptor governing transport behavior, while entropy of mixing (ΔS mix ) and mixing enthalpy (ΔH mix ) contribute secondary, mechanistically consistent effects. The proposed HE–ML framework offers a rapid approach for predicting thermoelectric properties and enabling data-driven compositional screening. • ML with HEA descriptors predicts EC, SC, and TC with high accuracy. • XGB yields the lowest RMSE test and highest R 2 test (>0.9) across all properties. • δ dominates transport behavior, with ΔS mix and ΔH mix providing secondary effects. • HEA–ML framework enables rapid and interpretable screening of TE materials.