Obtaining Minimum Error of Classification with Modified Fisher's Cutoff Point

Ogbonna Eric Nnamdi, Abam Ayeni Omini, Nsien Edwin Frank

Abstract: Fishers method of adjusting the cutoff point to obtain a balance on the total probability of misclassification are disadvantaged in the sense that they inflate the classification error in the presence of outliers. The proposed method of Modified Fishers Cutoff Point (MFCP) is aimed at addressing this problem. To accomplish this purpose, linear discriminant analysis was performed using Fishers technique and the modified approach. Misclassification errors were computed for both methods and the modified procedure was observed to have minimum error than the Fishers procedure. The results obtained in applying the cutoff point showed that the modified approach performs better than the Fishers linear discriminant cut off point. The application of this new method reduces the effect of outliers to the barest minimum.

Keywords: Cutoff point, outliers, misclassification, apper, and discriminant