The application of support vector machines to the identification of materials attributes

Machine learning and data mining tools have not been widely used in materials science. In this work we examine how a machine learning methodology, support vector machines (SVM), can be used to predict or detect patterns of behavior in the properties of materials. In this paper we demonstrate the use of SVM classification as a tool to classify whether materials can be high temperature superconductors or not. The eventual goal is to use this classification capability to design new materials with the desired physical properties. Cross-validation demonstrates that SVM can predict superconductivity with high accuracy. Thus SVM and other related machine learning are very promising tools for enhancing the search for new materials.