Non-Parametric Discriminant Analysis
Non-Parametric Discriminant Analysis
复制标题
DOI：
10.1007/978-3-642-83520-9_60
复制
发表时间：
1988
期刊：
影响因子：
--
通讯作者：
N. Lack
中科院分区：
文献类型：
--
作者：
N. Lack
最近有一篇关于使用分类或混合预测变量的判别分析的扩展文献。在其中几份报告中，定性判别法与经典线性判别法进行了比较。泰特灵顿(1)对不同的方法作了最全面的描述。不太常用的方法有“最简单最一般的判别法”、摩尔方法(2)和Goldstein和Dillon提出的分布距离法(3)。虽然质心法适用于变量的任何组合，但分布距离法最容易适用于二元预测变量。这里只考虑两个群体之间的歧视。将算法推广到两个以上组的情况不存在理论问题，而是在推导分配规则时只涉及添加代数。在下文中，这两种不太流行的定性方法将与经典的线性判别分析在二元预测变量和两个总体之间的区分的情况下进行比较。
There has been a recent expanding literature on the use of discriminant analysis with categorical or mixed predictor variables. In several of these reports qualitative discriminant approaches have been compared with classical linear discrimination methods. A most comprehensive account of different methods is given by Titterington (1). Less common approaches include thecentroid method, ‘the simplest most general discriminant rule’, Moore, (2) and thedistributional distancemethod as proposed by Goldstein and Dillon (3). Whilst the centroid method is suited to any combination of variables the distributional distance method is most readily applicable to binary predictor variables. Only discrimination between two groups will be considered here. Extension of the algorithms to the case of more than two groups presents no theoretical problems but rather involves only added algebra in the derivation of the allocation rules. In the following these two less popular qualitative methods will be compared with classical linear discriminant analysis for the case of binary predictor variables and discrimination between two populations.