What if my tree is too large or too complex?
If CART determines that the optimal tree has a very large number of nodes and therefore is too complex for practical use or easy intelligibility, two solutions are possible.
SOLUTION 1: Pick a Smaller Tree
CART is designed to identify a set of candidate predictive trees complete with honest estimates of costs and standard errors. There is no reason not to decide to accept a higher error rate in exchange for a simpler tree, so long as you remain aware of the costs of the simplification.
SOLUTION 2: Parametric Model
Consider fitting a parametric model using CART to select variables and possible interactions. We have often used CART regression trees to partition a sample into subsets on which separate linear models are fit. CART can assist in specifying a switching regression.Steinberg, Dan and Phillip Colla. CART—Classification and Regression Trees. San Diego, CA: Salford Systems, 1997.