Evaluation of an adaptive discriminant procedure

This paper reports the results of a simulation study comparing Fisher's Linear Discriminant Function, Smith's Quadratic Discriminant Function, a nonparametric Nearest Neighbor approach, a linear programming approach, and a new adaptive statistical method for solving the discriminant problem. The study examines the two-group discriminant problem with four variables, of which two are continuous and two are discrete. The analysis is based on the rate of misclassification using each method. The results indicate that the adaptive method is an effective alternative to existing methods and that the adaptive philosophy of using the training sample to identify which of several methods should be applied to the validation sample merits further study.