Model Explainability

Model Explainability Techniques

Default Feature Importances

Tree-based models like Random Forest or XGBoost have built-in feature importance calculations, but there are important considerations to take into account when using these:

Beware Default Random Forest Importances

Important reading for anyone using linear model coefficients or tree-based feature importance from Terence Parr et al.

Oct 20, 2018

SHAP

If you do a web search for model explainability, one of the first results will likely include SHAP. If you are wondering about when to use SHAP or, even if you use it already and want to read some important discussion around it, there is a group of three posts here that provide useful food for thought:

SHAP Is Not All You Need

Three blog posts about SHAP and alternatives to consider from Christoph Molnar and Giles Hooker

Jun 18, 2024