Discrimination with categorical variables

被引：0

作者：

Buttrey, SE ^{[1
]}

机构：

[1] USN, Postgrad Sch, Dept Operat Res, Monterey, CA 93943 USA

来源：

DIMENSION REDUCTION, COMPUTATIONAL COMPLEXITY AND INFORMATION | 1998年 / 30卷

关键词：

discrimination; classification; optimal scaling; categorical variables;

D O I：

暂无

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

A new method for discrimination when all or most predictor variables are categorical is given. In this technique the class labels and categorical values are simultaneously mapped into red numbers so as to minimize a particular sum of squares. In the case of C classes it turns out that each class can be represented in a space whose dimension is at most C - 1. Observations can also be located in this space. This leads to a discrimination rule requiring the computation of at most C - 1 weighted Euclidean distances. For some well-known data sets, this technique produces accuracy competitive with any other classifier.

引用

页码：550 / 555

页数：6