典型相关分析的延拓研究

An Extension of the Canonical Correlation Analysis

在典型相关分析中,求得典型相关变量的表达式并没有全部完成任务,例如需要确定典型相关变量的个数和变量选择。针对典型相关变量的个数问题,发现了常用的卡方检验和冗余分析方法的不足,进而提出了一种新的算法。针对原始变量的选择问题,提出了三种可能的路径。最后利用人体尺寸数据对相关结论进行了验证。

The canonical correlation analysis is not finish when we get the expression of the canonical correlation variables.We need decide the number of the canonical correlation variables and the optimal combination of the original variables.For the first question,we show how to use Chi-square test or redundancy analysis to solve it and find the deficiencies of these methods.Then,we give a new algorithm to the first question.For the second question,we discuss three different ways to find the optimal combination.Finally,we use these algorithms in the study of relationship between the length of human upper and lower limbs.