Validating Classification Trees
K. D. Wernecke, K. Possinger, G. Kalb, Julia Stein · Biometrical Journal · 1998
Nonparametric statistical classification methods that work without specific requirements to the distribution of the underlying data play an important role especially in medical applications. Such a method is that of “Classification and Regression Trees” (CART), already suggested by Sonquist, 1970 and improved later on by Breiman et al., 1984. Of essential importance for the medical practice is the assessment of the validity of the tree obtained. In this paper resampling-methods are used to this purpose. With the construction of a classification tree resulting from the validation process we not only get a valuation of the stability of the tree construction but also a tree to be applied in clinical practice.