[16a-K505-1~9] 23.1 Joint Session N "Informatics"

11:00 AM - 11:15 AM JST(2:00 AM - 2:15 AM UTC)

[16a-K505-8] Establishing Explainable Neural Network for Predicting Physical Properties of Functional Polymer

〇(D)YinKan Phua¹, Tsuyohiko Fujigaya^1,2,3, Koichiro Kato^1,2,4 (1.Grad. Sch. of Eng., Kyushu Univ., 2.CMS, Kyushu Univ., 3.I2CNER, Kyushu Univ., 4.RIIT, Kyushu Univ.)

Keywords:

Fuel cell,Explainable,Machine learning

Neural networks (NNs) are widely utilized in materials informatics due to their high predictive accuracy; however, their low explainability limits their application primarily to the prediction of properties for novel materials. Enhancing explainability could open new possibility of deriving novel materials design guidelines. In this study, we developed an explainable NN for functional polymers that allows users to analyze the importance of each explanatory variables by leveraging SHAP values. By interpreting the key explanatory variables identified through SHAP value analysis in a chemical context, we aim to extract novel materials design guidelines that could accelerate the research of functional polymers.