Machine learning modeling of superconducting critical temperature

Connection between superconductivity and chemical and structural properties of materials is the key to understanding the mechanisms of superconductivity, and yet finding this connection is major experimental and theoretical challenge. We have developed several machine learning methods for modeling the critical temperatures T_c of the 12,000+ known superconductors available via the SuperCon database. Materials are first divided into two classes based on their T_c values, above and below 10 K, and a classification model predicting this label is trained. The model uses coarse-grained features based only on the chemical compositions. It shows strong predictive power, with out-of-sample accuracy of 92%. Separate regression models are developed to predict the values of T_c for cuprate, iron-based, and low- T_c compounds. These models demonstrate good performance, with learned predictors offering insights into the mechanisms behind superconductivity in different families. We combined the classification and regression models into a single pipeline and employed it to search the entire Inorganic Crystallographic Structure Database for potential new superconductors. We have identified 35 oxides as candidate materials.