Machine Learning Classification Model for Screening of Infrared Nonlinear Optical Crystals

Infrared nonlinear optical crystals are indispensable functional materials in modern laser science. In order to avoid the inherent defects of traditional materials such as two-photon absorption and low laser damage threshold, there is still an urgent need to find infrared nonlinear optical crystals with excellent properties. In this paper, we combine the machine learning (ML) algorithm with the features that only contain crystal chemical components, train the nonlinear optical crystal database containing more than 400 materials, and optimize the ML classification model. The random forest classifier with synthetic minority oversampling technique (SMOTE) sampling shows good classification ability. An external test is used to test the generalizability of the model, and the result shows that the model meets the requirements. The combination of the ML classification model and high-throughput computing is helpful to accelerate the discovery of new crystals.