#permutation-importance

Question 1: Single choice

A random forest predicts taxi fares. You multiply one feature, the absolute change in latitude, by 1000 (as if changing its unit), retrain, and recompute permutation importance on the validation data. What happens to that feature's permutation importance?

It stays about the same; tree splits depend on order, not scale1
It grows about 1000-fold, because the feature's values are 1000 times larger2
It shrinks, because shuffling large values disturbs the predictions less3
It cannot be computed until the feature is standardized again4

#permutation-importance

Timed test