An Asymptotic Unbiased Technique for Estimating the Error Rates in Discriminant Analysis

A new technique for estimating the error rates in discriminant analysis is suggested. The estimates obtained using this technique are asymptotically less biased than existing estimates. The performance of this technique on the basis of criteria other than that of bias is studied using Monte Carlo methods to simulate practical situations and also the criterion of asymptotic mean square error. It is concluded that the all-round performance of the proposed technique is comparable to that of any available technique.