材料研究中的可解释机器学习

材料研究中的可解释机器学习

王冠杰, 刘盛咸, 周健, 孙志梅

Explainable Machine Learning in the Research of Materials Science

WANG Guanjie, LIU Shengxian, ZHOU Jian, SUN Zhimei

表 1 XML方法评估结果

Table 1 Evaluation results of XML methods

Classification	XML Method	Integrity	Expressiveness	Transparency	Portability	Complexity	Understandability
Interpretable methods from internal model structure	Linear regression Logistic regression	High	High	High	Low	Simple	Easy
Interpretable methods from internal model structure	Linear regression Logistic regression	High	Low^a	Low	Not portable	Middle	Easy
	Decision tree	High	High	High	High^b	Middle	Easy
	K nearest neighbors	High	High	High	Not portable	Simple	Easy
Interpretable methods for external model evaluation methods	Feature importance ranking and partial dependence plots	Low	Limited^c	Low	High^d	Simple	Easy
	Feature importance ranking and partial dependence plots	Low	Limited^c	Low	High^d	Simple	Easy
	Shapley additive explanation(SHAP)	Low^e	Extremely high for single predictions	Low	High	Middle	Easy
	Shapley additive explanation(SHAP)	Low^e	Extremely high for single predictions	Low	High	Middle	Easy
	LRP	Low	High	High	High	Complex	Easy
	LRP	Low	High	High	High	Complex	Easy
	CAM	Low	High	High	High	Complex	Easy