Heteroscedastic Discriminant Analysis with Expert Systems and NMR Applications
Edward J. Dudewicz, Vidya S. Taneja · American Journal of Mathematical and Management Sciences · 1989
SYNOPTIC ABSTRACTThe problem of “discriminant analysis,” i.e. of providing a method for classifying new observations into one of k≥2 populations, has been studied by many authors since the area was opened by Fisher in 1936. The probabilities of misclassification are of key interest, and present procedures yield only hopeful approximations to them when population parameters are unknown (as is the case in practice). In this paper we give a new Heteroscedastic Discriminant Analysis Procedure which provides precise estimation of the misclassification probabilities. The results are being incorporated into ESS™, The Expert Statistical System, and are illustrated on NMR data from the human brain.