Interpretable Machine Learning Workflow for Evaluating and Analyzing the Performance of High-Entropy GeTe-Based Thermoelectric Materials

Abstract

To guide the development of high-performance thermoelectric materials, it is essential to design appropriate material compositions and temperature environments. This study focuses on analyzing the properties of high-entropy GeTe-based thermoelectric materials under different temperature environments and chemical compositions using an interpretable machine learning workflow. First, an experimental dataset from previous research on thermoelectric materials is constructed, and descriptors based on atomic features are established. Feature selection techniques, such as Pearson correlation, univariate feature selection, and exhaustive feature selection, are applied to select relevant features. The selected features are then used in conjunction with the target variable, the ZT value, for training and testing. The effectiveness of the training is demonstrated by comparing the performance on the testing set and using cross-validation to identify the best machine learning model. Furthermore, the SHAP (SHapley Additive exPlanations) method is employed to interpret the best model. Through global interpretability, analysis of interactive variables' contributions to the target variable, and local interpretability methods, material selection, and performance optimization are carried out. The results reveal that temperature has the greatest impact on the ZT value of thermoelectric materials, while the effects of molar volume and electronegativity sequentially diminish. By utilizing interpretable machine learning methods, we are able to predict and optimize the performance of thermoelectric materials based on known samples. This not only facilitates the evaluation and prediction of the thermoelectric properties of materials but also provides a comprehensive research workflow design approach for guiding the selection and development of high-performance GeTe-based thermoelectric materials.

Machine Learning Accelerating the Condition Screening of Ceftriaxone Sodium Anaerobic Co-Metabolic Degradation

“…SHAP analysis is a marginal analysis mechanism . Its core principle is to calculate the marginal contribution of features to the model output, then explain the “black box model” on both the global and local levels .…”

“…SHAP analysis is a marginal analysis mechanism. 29 Its core principle is to calculate the marginal contribution of features to the model output, then explain the "black box model" on both the global and local levels. 30 SHAP builds an additive explanation model, all features are regarded as "contributors", and the mechanisms of influence behind different influencing factors on the target value are explained by deconstructing the contribution value.…”

A comprehensive review of entropy engineered GeTe: an antidote to phase transformation

“…From the viewpoint of transport of charge, an enhancement of either s–p hybridization or ionicity of the bond augments the isotropic nature of the light valence band (L-band), which eventually decreases its and thus makes it superior among group IV–VI chalcogenides. 129–135 Fig. 8(b) demonstrates that although the cubic structure (at elevated temperature) has high symmetry in comparison to the rhombohedral structure at low temperature, as far as thermoelectricity is concerned, slight symmetry breaking is preferred as it leads to high band degeneracy.…”

“…The methodology involves constructing the dataset from experiments based on preceding research on thermoelectrics and subsequently descriptors are established depending on the atomic features. 131 To choose pertinent features, some of the feature selection mechanisms, viz. , Pearson correlation, exhaustive and univariate feature selection, are considered.…”

High-entropy materials for thermoelectric applications: towards performance and reliability

“…Another family of HEMCs that has also garnered attention to explore their TE properties is the HE tellurides. 210,217 The first investigated system in this class was the GeTe-based TEMs, one noteworthy example is the Ge 0.61 Ag 0.11 Sb 0.13 Pb 0.12 Bi 0.01 Te HE material based on the (Ge/Ag/Sb/Pb)Te system reported by Jiang et al 42 The incorporation of several alloying Ag, Sb, and Pb elements into Ge sites has increased the solubility (by ∼10%) compared to the solubility in the presence of the individual elements. This leads to an improved symmetry in the crystal structure, which is near to the cubic phase in the (Ge/Ag/Sb/Pb)Te system.…”