Clustered Feature Importance (Presentation Slides)
35 Pages Posted: 6 Mar 2020 Last revised: 28 May 2020
Date Written: January 29, 2020
A substitution effect takes place when two or more explanatory variables share a substantial amount of information (predictive power).
Under the presence of substitution effects, feature importance methods may not be able to determine robustly which variables are significant.
This presentation discusses the Clustered Feature Importance (CFI) method, which is robust to linear as well as non-linear substitution effects.
Keywords: machine learning, feature importance, permutation importance, mean decrease accuracy
JEL Classification: G0, G1, G2, G15, G24, E44
Suggested Citation: Suggested Citation