Estimating Misclassification Probabilities for the Quadratic Discriminant Function by Saddlepoint Approximation

摘要

In this paper, the cumulant generating function (CGF) of the quadratic discriminant function (QDF) is obtained. Since CGF of the QDF is known, we use saddlepoint density estimation for approximating QDF distribution. The defects and disadvantages of approximating QDF distribution were determined using simulation and the well-known Iris data. The finding results reveal suitable estimation of misclassification probabilities, in light of the saddlepoint approximation.