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

• DOI: 10.1021/acsami.2c15396
• Publication Date: 2022-12-06T14:27:36Z
• Authors: Yi Li; Jingzi Zhang; Ke Zhang; Mengkun Zhao; Kailong Hu; Xi Lin
• Keywords: Hyperparameter; Figure of Merit; Gradient boosting; Boosting; Root mean square; Data set; Feature Engineering; Feature (linguistics)

The figure of merit (zT) is a key parameter to measure the performance thermoelectric materials. At present, prediction zT values via machine leaning has emerged as promising method for exploring high-performance However, learning-based predictions still suffer from unsatisfactory accuracy, and this related size data set, hyperparameters models, quality data. In work, 5038 pieces materials were selected, several regression models generated predict values. This large set-driven light gradient boosting (LGB) model with 57 features performed an excellent achieving coefficient determination (R2) value 0.959, root mean squared error (RMSE) 0.094, absolute (MAE) 0.057, correlation (R) 0.979. Owing accuracy exceeds that most reported learning. "ME Lattice Parameter" was verified important feature in prediction. Furthermore, nine potential candidates screened out among one million study solves problem set size, adjusts uses engineering improve quality, provides efficient strategy perform wide-ranging screening