[求助]采用聚类分析时，如何确定聚类的数目？

BIC?怎么没有图啊??

上传了，也提示上传成功了。结果却没有显示，不知为什么。因为是公式，这里打不出来

是啊，看不见，能否存成图片格式再传一下，谢谢拉

我也来学习学习

解释的相当详细，谢谢各位大侠的论述。

前面好多方法，意见．补充一点：多试几次，再检验几次．直到定量与定性都达到你的目的～～｀

一点点的试了，只有这样。

The NUMCLUSTERS subcommand specifies the number of clusters into which the data will
be partitioned.

AUTO Automatic selection of the number of clusters. Under AUTO, you may specify
a maximum number of possible clusters. TWOSTEP CLUSTER will search for
the best number of clusters between 1 and the maximum using the criterion
that you specify. The criterion for deciding the number of clusters can be
either the Bayesian Information Criterion (BIC) or Akaike Information Criterion
(AIC). TWOSTEP CLUSTER will find at least one cluster if the AUTO
keyword is given.

FIXED User-specified number of clusters. Specify a positive integer

Examples

TWOSTEP CLUSTER
/CONTINUOUS VARIABLES = INCOME
/CATEGORICAL VARIABLES = GENDER RACE
/NUMCLUSTERS AUTO 10 AIC
/PRINT SUMMARY COUNT.

TWOSTEP CLUSTER uses the variables RACE, GENDER and INCOME for clustering. Specifications
on the NUMCLUSTERS subcommand will instruct the procedure to automatically
search for the number of clusters using the Akaike Information Criterion and require the
answer to lie between 1 and 10.

===================================================================
TWOSTEP CLUSTER
/CONTINUOUS VARIABLES = INCOME
/CATEGORICAL VARIABLES = RACE GENDER
/NUMCLUSTERS FIXED 7
/PRINT SUMMARY COUNT.
Here the procedure will find exactly seven clusters.

[此贴子已经被作者于2008-2-13 10:30:52编辑过]

个人认为SPSS的聚类只是数学方式的聚类，如果想使聚类真的有意义，还需要与数据的现实意义紧密结合，才能从多种聚类方案中选出真正可用的。

顶一下