Large Data Set-Driven Machine Learning Models for Accurate Prediction of the Thermoelectric Figure of Merit

Abstract

The figure of merit (zT) is a key parameter to measure the performance of thermoelectric materials. At present, the prediction of zT values via machine leaning has emerged as a promising method for exploring high-performance materials. However, the machine learning-based predictions still suffer from unsatisfactory accuracy, and this is related to the size of the data set, the hyperparameters of models, and the quality of the data. In this work, 5038 pieces of data of thermoelectric materials were selected, and several regression models were generated to predict zT values. This large data set-driven light gradient boosting (LGB) model with 57 features performed with an excellent accuracy, achieving a coefficient of determination (R 2) value of 0.959, a root mean squared error (RMSE) of 0.094, a mean absolute error (MAE) of 0.057, and a correlation coefficient (R) of 0.979. Owing to the large size of the data set, the prediction accuracy exceeds that of most reported zT predictions via machine learning. The “ME Lattice Parameter” was verified as the most important feature in the zT prediction. Furthermore, nine potential candidates were screened out from among one million pieces of data. This study solves the problem of the data set size, adjusts the hyperparameters of the models, uses feature engineering to improve data quality, and provides an efficient strategy to perform wide-ranging screening for promising materials.

Optical property dataset of inorganic phosphor

“…Recently, various groups have reported experimental physical property prediction using a chemical formula-based feature alone 22 – 27 . In addition to chemical formula-based features, if additional information related to physical property measurements is used as a feature to generate a prediction model, further improvement of the prediction model can be expected.…”

Accurate and efficient machine learning models for predicting hydrogen evolution reaction catalysts based on structural and electronic feature engineering in alloys

“…7,8 However, trial-and-error experiments and density functional theory (DFT) calculations are challenging due to their time-consuming processes to explore the entire chemical composition of alloys. 9 Over the past decade, datadriven machine learning (ML) models have provided alternative approaches to material design in various research fields, such as superconducting materials, 10,11 thermoelectric materials, 12,13 and electrochemical catalysts. 14,15 Moreover, ML models and feature analysis were utilized to predict the ΔG H* values of alloys, providing new perspectives and accelerating the exploration of potential HER catalysts.…”

Deep Generative Model for Inverse Design of High-Temperature Superconductor Compositions with Predicted Tc > 77 K

“…In recent years, artificial intelligence (AI) has made great progress in the discovery of new materials. − In order to speed up the hunt for novel high T c superconductors, researchers are equipped with machine learning (ML) to screen potential candidate new materials. , Several T c predictors were first constructed in superconductor databases such as SuperCon. Stanev et al, for instance, employed random forest approach together with element descriptors to predict superconductor T c , and their coefficient of determination R 2 was 0.88; despite the discovery of 35 superconductors from ICSD, none of them is predicted to have T c excess 40 K. , Zeng et al successfully predicted the T c of superconductors by using the atom table convolutional neural network (ATCNN) model with an R 2 of 0.96, which can directly learn the experimental characteristics from the element descriptors .…”