The Greedy Boruta Algorithm: Faster Feature Selection Without Sacrificing Recall

The Boruta algorithm is a powerful "all-relevant" feature selection method, but its rigorous statistical tests can be computationally expensive and slow to converge. A proposed modification, the Greedy Boruta algorithm, solves this by dramatically relaxing the confirmation rule. This new approach immediately confirms any feature that outperforms its randomized "shadow" version even once, rather than waiting for statistical proof over many iterations. As a result, Greedy Boruta offers guaranteed and significantly faster convergence while provably maintaining or improving the recall of important features.