DISTRIBUTION OF ERRORS OF MISCLASSIFICATION FOR THE LINEAR DISCRIMINANT FUNCTION (A CASE OF EDGEWORTH SERIES NON NORMAL DISTRIBUTION)

Abstract

In this paper, the discrimination and classification problem associated with the persistent non normal distribution has been studied. Sampling  from non normal distribution is assessed through the  distribution of errors of misclassification in respect of Edgeworth Series Distribution (ESD) which is restricted to asymmetry. The effects of applying a normal classificatory rule (ND) when the distribution is ESD by empirical approach is examined  by comparing the errors of misclassication  for ESD with ND using small sample sizes at every level of skewness factor. The empirical results obtained show that the normal procedure is sturdy against departure from normality. This is evident from the total probabilities of misclassification that are not greatly affected by the skewness factor.

Keywords: Normal Distribution, Classificatory Rules, Apparent Probability of Misclassification,

Skewness Factor and Optimum Probabilities of Classification.