La distribuzione del tasso di errore apperente nei criteri di classificazione ad albero

Abstract

The aim of this paper is to study the distribution of the apparent error rate in binary trees structured classifiers. This is the most commonly used estimator in discriminate analysis. After a brief review on binary trees and on the estimation of the error rate, we describe the application of bootstrap sampling to the above problem. Then, we present a simulation experiment concerning the classic problem of discrimination between two species of Iris.