A machine learning approach to the prediction of the dispersion property of oxide glass

Abstract

In this study, we built a model for predicting the optical dispersion property of oxide glasses via machine-learning techniques such as kernel ridge regression, neural networks, and random forests. The models precisely predicted the optical property. Based on the predictions for glasses with doped oxides, we prepared new glasses in our laboratory. The experiments agreed well with the predictions made using kernel ridge regression and neural networks but not with those made using random forests. The results of this study demonstrate that the data-driven approach is a promising route for new material design.