Machine-learning guided prediction of thermoelectric properties of topological insulator Bi2Te3-xSex

Thermoelectric materials play a pivotal role in harnessing waste heat and converting it into valuable electrical energy, addressing energy sustainability challenges. This study introduces an innovative methodology to predict essential thermoelectric properties-thermal conductivity (kappa), electrical conductivity (sigma), Seebeck coefficient (S), and the figure of merit (ZT)-solely from the chemical formula of materials. Employing advanced machine learning (ML) techniques, including random forest, gradient boosting regression, XGBRegressor, and Neural Network, we developed a robust predictive model utilizing a diverse dataset of thermoelectric compounds. Notably, random forest exhibits outstanding predictive performance, boasting R-2 values of 0.91, 0.95, 0.95, and 0.90 for kappa, sigma, S and ZT, respectively. While testing the prediction competency of thermoelectric parameters of Bi2Te1-xSex using a random forest model, the model provides a very consistent quantitative prediction with experimental kappa, sigma, S and ZT. Furthermore, the kappa, sigma, S and ZT of Bi2Te2Se were calculated using the first principles density functional theory and Boltzmann transport equation to compare the corresponding ML-predicted thermoelectric properties. Although the order of theoretical values of kappa, sigma, S and ZT of Bi2Te2Se is consistent with the room temperature ML prediction, the temperature-dependent theoretical value of kappa, sigma, S and ZT of Bi2Te2Se shows a deviation from the ML-prediction values as the model is trained with the experimental data. The findings highlight the superiority of classification-based models in capturing complex patterns. By leveraging chemical composition as the exclusive input, our streamlined approach eliminates the need for extensive laboratory experiments. This research significantly propels the advancement of high-performance thermoelectric materials, offering an efficient pathway for exploration and optimization, thus revolutionizing the field of materials science. Moreover, it opens avenues for accelerated discovery and innovation in energy conversion technologies.