What is a limitation of permutation importance?

Question: What is a common limitation of permutation feature importance? A) It only works for unsupervised learning B) It cannot be used after model training C) It can give misleading results when features are strongly correlated D) It requires the model to be linear Correct: C Explanation: If two features are highly correlated, permuting one feature may not greatly reduce performance because the model can still use the other correlated feature. This can underestimate the importance of correlated predictors. Topic: advanced ML / interpretability / feature importance