CART relies on the concept of prunning to create a sequence of nested models as final model candidates. Independent testing and cross validation is subsequently used to identify the optimal tree with respect to overall model performance.