XAI, Considerations of SHAP Values

## Détails

Title :
Considerations of SHAP Values in the settings of Correlated and Causal Features
Abstract:
SHAP values, designed to allocate the contribution of each feature to the prediction made by any model, form an important and a very widely used framework for machine learning interpretability. However, special care must be taken when features are correlated or even with causal relationships. In this talk, I will discuss the following topics with concrete examples and use cases:

• How SHAP values can be misleading or even not correct when features are correlated.
• How SHAP values can be incorrect when there are causal relationships between features.
• An overview of causal SHAP as a solution for causality.

Speaker : Shuyang XIANG, Lead Data Scientist | Mathematician | Blogger & Educator
Blog posts:

• https://medium.com/towards-data-science/kernelshap-can-be-misleading-with-correlated-predictors-9f64108f7cfb
• https://medium.com/towards-data-science/casual-shap-values-a-possible-improvement-of-shap-values-4d4d62925b71