At what stage are random features selected in a random forest: at each split or for the entire tree?

In the random forest algorithm, the random selection of features occurs at each split (node) of the tree, not just once for the entire tree.

This means that when constructing each node of the tree, a subset of features is randomly selected from the entire set of features, and the best one for splitting is chosen from this subset. This approach increases the diversity of trees in the forest, reduces correlation between them, and improves the model's generalization ability.

If features were selected only once for the entire tree, then all splits within that tree would be limited to the same subset of features, which would reduce the diversity and effectiveness of the ensemble.