BOOTSTRAP PERCENTILE CONFIDENCE INTERVALS FOR ACTUAL ERROR RATE IN LINEAR DISCRIMINANT ANALYSIS

In this study, bootstrap percentile confidence intervals for the actual error rate have been obtained with respect to the bootstrap estimated values of the estimators D, DS, L, LS, O, OS and All. Estimated values for these estimators have been obtained by a simulation study for two multivariate normal populations with different mean vectors and a common variance-covariance matrix. Moreover, traditional confidence intervals for the actual error rate were obtained based on the analytic form of the estimator All. In general, bootstrap percentile intervals are narrower than the traditional intervals obtained with respect to the analytic form of the estimator M. Narrower intervals are desired which will be relatively consistent with the parameter. It is shown that some statistical properties of the estimators can be obtained by the bootstrap method.