A Generalizable Machine Learning Framework for Half-Heusler Thermoelectrics
The study developed a robust machine learning workflow for predicting thermoelectric figure of merit (zT) in half-Heusler materials. It improved model generalizability by using PCA-based data splitting to ensure unbiased training and testing. The framework integrates feature selection, Bayesian optimization across three models, and physical insight from symbolic regression to identify key material properties.