-,SYMPTOTIC EFFECTS OF THE MAHALANOBIS DISTANCE ON THE LINEAR ,. CLASSIFICATION FUNCTION

Abstract

The performance of the Linear Classification Function has been investigated in previous works. We use a Monte Carlo study to investigate the asymptotic effect of the Mahalanobis distance (1:12) when observa tions are from multivariate normal populations. Results show that increasing the sample size does not trans late into improved performance of the function once an optimal size is exceeded for all values of 8 ( ) and the total error rate displays after this value. There is a reduction in variation (as expected) and the error rates approach the value of the standard deviation after d=5, and overlapping at d=7.