The decision tree algorithms build an initial tree from the training data first, and then prune this tree to increase its accuracy on test data.