SHAP scores can assign zero importance to a relevant feature and nonzero importance to an irrelevant feature, even for Lipschitz-continuous and arbitrarily differentiable regression models.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SHAP scores fail pervasively even when Lipschitz succeeds
SHAP scores can assign zero importance to a relevant feature and nonzero importance to an irrelevant feature, even for Lipschitz-continuous and arbitrarily differentiable regression models.