分类回归树中分裂法则的相合性质

THE CONSISTENCE OF SPLITING RULE IN CLASSIFICATION AND REGRESSION TREES

讨论度量误差对分类树回归结果的影响，证明了对于某些常用的分裂法则来说，度量误差的存在会影响类边界估计的相合性.在一种简单情况下，提出了一个分裂法则，在该法则下，度量误差的存在不会影响类边界的相合性.

Classification and regression trees is a kind of nonparametric statistic method based on iteration. This article discusses the influence of measure error on classification and regression tree's results, and shows that the measure error will affect the consistence of estimate of classification's border for some typical use. Under a simple case, we present a new splitting rule by which the measure error does not affect the consistence.