Predicting the thermal conductivity of Bi2Te3-based thermoelectric energy materials: A machine learning approach

Abstract

Bi2Te3-based materials are remarkable thermoelectric renewable energy harvesters. The measurement of their thermal conductivity (κ) is a critical phase toward the realization of the material's energy conversion efficiency. However, the experimental techniques involved in the measurements of κ, particularly for thin films, are incredibly challenging. Herein, we introduce a pioneering technique using support vector regression and decision tree regression machine learning models where the values of κ can be predicted based on the structural crystal lattice constants of the material and its electrical properties. A decision tree regression (DTR) and support vector regression (SVR) models using both radial basis function (RBF) and polynomial kernels were developed. The performance of the models was evaluated based on the correlation coefficient (CC) between the predicted and actual values of κ, R2 values, mean absolute error (MAE), and mean square error (MSE). Our results revealed that the DTR outperforms the SVR models in estimating the values of κ with CC of 98.7% and R2 of 97.5% for the testing phase. The models were validated by solving some real-world problems such as predicting the thermal conductivity of transition metal-doped Cu–Bi2Te3, effects of doping non-metals (Se–Bi2Te3), and the role of toxic elements (Pb–Bi2Te3) on the values of κ. The model was further employed to investigate the effects of pulsed laser deposition substrate temperature on the thermal conductivity of Bi2Te3. The performance of the models in predicting the thermal conductivity of Bi2Te3-based materials makes it a useful tool for thermoelectric energy research.