A NEW DISCRIMINANT TECHNIQUE - BAYES-FISHER DISCRIMINATION

In this paper, a new discriminant technique, called Bayes-Fisher (B-F) discrimination, is presented to deal with the discrimination problem involving k distinct populations when the prior probabilities p(i) of the k distinct populations G(i) (i = 1, ..., k) are known while the distributions of G(i) are unknown. The relationship between the B-F discriminant and other discriminants is investigated. It is shown that B-F discrimination is a reasonable extension from Fisher's when the prior probabilities are known. A practical example is given.