MONTE CARLO EVALUATION OF ERROR RATE ESTIMATORS IN DISCRIMINANT ANALYSIS UNDER MULTIVARIATE NORMAL DATA

Abstract

This paper is concerned with the problem of estimating the error rate in two-group discriminant analysis. Here, behaviour of 19 existing error rate estimators are compared and contrasted by mean of Monte Carlo simulations under the ideal condition that both parent populations are multivariate normal with common covariance matrix. The criterion used for comparing those error rate estimators is sum squared error (SSE). Five experimental factors are considered for the simulation, they are the number of variables, the sample size relative to the number of variables, the Mahalanobis squared distance between the two populations, dependency factor among variables, and the degree of variation among the elements of the mean vector of the populations. The result of the simulation shows that there is no estimator performing the best for all situations. However, on overall, the Finite Mixture Balanced bootstrap estimator (FMB) proposed by Mangku (2007) is the best estimator.