Fisher’s Linear Discriminant Function Versus Logistic Regression: An Argument for Separate Group Accuracies and Validation

• In Event: Machine Learning Methods in Educational Research

Abstract

The purpose of this paper is to describe a method that compares the cross-validated classification accuracies of Fisher’s Linear Discriminant Function and Logistic Regression for an input data set. A computer program that implements the method is included. Given discrepant simulation results in comparing the accuracy of these methods, no generically best method can be advocated. Thus, the proposed procedure will allow researchers to examine the results of these alternate approaches for an input data set to judge their relative accuracies by both Leave-One-Out and Bootstrap cross-validation. Consideration of separate-group accuracies is advocated, and the effect is illustrated.

Authors

• John D. Morris, Florida Atlantic University

• Mary G. Lieberman, Florida Atlantic University

• Maria D. Vásquez-Colina, Florida Atlantic University