Machine learning modeling of superconducting critical temperature.

Superconductivity has been the focus of enormous research effort since itsdiscovery more than a century ago. Yet, some features of this unique phenomenonremain poorly understood; prime among these is the connection betweensuperconductivity and chemical/structural properties of materials. To bridgethe gap, several machine learning schemes are developed herein to model thecritical temperatures ($T_{\mathrm{c}}$) of the 12,000+ known superconductorsavailable via the SuperCon database. Materials are first divided into twoclasses based on their $T_{\mathrm{c}}$ values, above and below 10 K, and aclassification model predicting this label is trained. The model usescoarse-grained features based only on the chemical compositions. It showsstrong predictive power, with out-of-sample accuracy of about 92%. Separateregression models are developed to predict the values of $T_{\mathrm{c}}$ forcuprate, iron-based, and "low-$T_{\mathrm{c}}$" compounds. These models alsodemonstrate good performance, with learned predictors offering potentialinsights into the mechanisms behind superconductivity in different families ofmaterials. To improve the accuracy and interpretability of these models, newfeatures are incorporated using materials data from the AFLOW OnlineRepositories. Finally, the classification and regression models are combinedinto a single integrated pipeline and employed to search the entire InorganicCrystallographic Structure Database (ICSD) for potential new superconductors.We identify more than 30 non-cuprate and non-iron-based oxides as candidatematerials.