Iteratively train random forests and run statistical tests to classify important/unimportant features.
Boruta is a heuristic. We run it multiple times, checking that the feature selection is stable. We check that the number of random forest trees is large enough to generate stable outputs.
Bias-Variance Tradeoff