[3E5-GS-2] Machine learning: Explainable AI (2)

[3E5-GS-2-04] Verification of Usefulness of SHAP values in Interpretation of Decision Tree Models

〇Shuho Yoshida¹, Yuki Tajima¹, Yusaku Imai¹ (1. Dentsu Digital Inc.)

Keywords:Interpretability, SHAP

In creating a predictive model by machine learning, it is important to provide interpretability as to which feature has contributed to model learning in what form in interpreting data and improving the model. Especially in the marketing industry, in addition to providing a mechanism to model and predict appropriate KPIs by machine learning, it is often required that get interpretability what variables influence KPIs and how they affect them. In recent years, what is called a SHAP value has been devised and attracted attention as an index for evaluating the contribution of the input features to model learning. In this paper, we show that SHAP values can roughly accurately evaluate the contribution of features to model learning in decision tree-based models often used for modeling table data.