Pruning

After Recursive Binary Splitting we may have a tree that overfits.

We grow some tree $T_0$ and prune it back. This pruned tree is a subtree. The pruned tree is a tree which produces the best (lowest) test error.