Thermoelectric MOFs and Machine Learning Databases for Prediction of Next Generation Materials

ABSTRACT Electrically conducting metal–organic frameworks (MOFs) have been studied over the last two decades and more recently research has gained interest on thermoelectric applications due to their ultra‐low thermal conductivity, which is attributed to their porous structure. Only a handful of intrinsically conducting MOFs and composite MOFs with conducting materials have been reported as thermoelectric materials. Here, thermoelectric MOFs have been comprehensively reviewed to prepare a dataset of 399 materials based on experimental observations. Each data point comprises 12 variables, which can be divided into two broad categories: the MOF's chemical and physical properties (composition, metal node, organic ligand, and crystal structure) and their thermoelectric properties (electrical conductivity, Seebeck coefficient, thermal conductivity, power factor, and figure of merit), which four are target variables of machine learning (ML) models. The dataset has been pre‐processed to streamline it. Six ML models, including Decision Tree, Random Forest, and Gradient Boosting, were trained on the dataset, and the Decision Tree‐based models can predict the electrical conductivity and Seebeck coefficient with high accuracy using the provided information in the dataset on thermoelectric MOF materials. The trained models using the thermoelectric MOFs dataset have been used to explore the thermoelectric parameters of 248 conductive MOFs, which have not yet been explored as thermoelectric materials. image