Exploratory Versus Decision Trees
Roberta Siciliano · COMPSTAT · 1998
Tree-structured methods using recursive partitioning procedures provide a powerful analysis tool for exploring the structure of data and for predicting the outcomes of new cases. Some attention is given to partitioning algorithms and this paper in particular recalls two-stage segmentation. The step from exploratory to decision trees requires the definition of simplification methods and statistical criteria to define the final rule for classifying/predicting unseen cases. Alternative strategies are considered, which result either in the selection of a method among the others or in the definition of a compromise among them.