[17p-A401-1~15] 23.1 Joint Session N "Informatics"

[17p-A401-8] SHAP analysis on machine-learned model using compositional descriptors

〇Kensei Terashima¹, Pedro Baptista de Castro^1,2, Akiko T. Saito¹, Takafumi D. Yamamoto¹, Ryo Matsumoto¹, Hiroyuki Takeya¹, Yoshihiko Takano^1,2 (1.NIMS, 2.Univ. of Tsukuba)

Keywords:machine learning, SHAP

SHAP is one of the emerging techniques for obtaining explainable AI (XAI) out of a constructed machine-learned model. In this talk, we will introduce our trial of applying this method to our model on magnetocaloric materials, and discuss how we can extract knowledge that the model acquired out of the regression process.