Avaliação da probabilidade de classificação incorreta em análises discriminantes para duas populações normais

Abstract

This work aimed to evaluate the performance of the Lachenbruch and Mickey (1968) method considering the modification proposed by Giri (2004) using Monte Carlo simulations and to measure and compare the misclassification rates with those obtained in the original method. Both methods original and modified are biased and have large biases in small samples and small biases with large samples. The biases of the two methods decrease with the increase of the Mahalanobis distance between the two populations. The original method is superior to the modified method, especially in small samples.