Distribution-free discriminant analysis

Abstract

Discriminant analysis is typically defined as the task of using predictor variables (such as blood chemistry variables) to predict a categorical response variable (such as whether the individual is diabetic). Although there are many good approaches, improvements are continually sought. Here, a non-parametric (distribution-free) discriminant analysis module is described. The module uses kernel density estimation to estimate the probability density for each class, and allows continuous, categorical, and ordered categorical predictors. Performance results on both real and simulated data sets and comparisons to other methods are provided. In some cases, this freely-available module performs better than the other methods. Nearly all cases can benefit from the application of multiple methods.