2021 Volume 20 Issue 1 Pages 14-21
In this study, we constructed machine learning models for predicting the relative permittivity (ε) and dielectric loss tangent (tanδ), which are important for the industrial application of thermosetting resin composites, using our own experimental data. We adopted a wide range of methods, including gradient boosting decision tree (GBDT) algorithms, which have been attracting attention in recent years, for the construction of machine learning models. Among the constructed models with multiple methods, we extracted models that satisfy the coefficient of determination R2CV > 0.8 at the time of cross-validation in the training data set. Furthermore, we selected the model in which the values of RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) were small in the training data set and could predict the physical properties more quantitatively and evaluated them in the test data set. As a result, we obtained a machine learning model in which RMSE and MAE can predict physical properties on the order of 10−1 to 10−2 for the mean values of ε and tanδ, respectively. From this result, we have demonstrated for the first time that the approach by MI (Materials Informatics) is effective even for thermosetting resin composites and that quantitative property prediction is possible. We expect that the development period will be shortened and promoted using this developed MI approach.
