Cluster Analysis Based On Correlation Matrices And Mixed Exponential Distribution

Cluster analysis is an important statistical method, and it is used widely. Thispaper describes two different situations to cluster.First, correlation matrices are used to cluster. The relations of multidimensionaldata are described by the correlation matrices. Then data are clustered throughtransforming correlation matrices. A method named Concor is discussed and is carriedout by iterating correlation matrices. Some relative properties of the method arepresented, and the Matlab program is introduced into the text. Fixed matrices based onConcor method are put forward when the rank is three, four and five.In the next place, data are supposed to have mixed distribution. In the systemassembled by several exponential distribution elements, it's helpful to present thesystem's distribution function for subtler study of the system life. This paper makes useof EM algorithm to calculate the MLE. And it use Gibbs sampling to obtain the MLEand posterior modes evaluations of mixed exponential distribution. The data of systemlife are clustered through taking in discriminated variables. However, in the practicethe number of the mixed distribution is unknown. In view of this situation we find thesolution which adopts the idea of reversible jump. So the number of mixed distributioncan be firstly estimated then the data can be clustered as the method put forwardbefore.At last, some common used cluster methods in gene expression data analysis areadvanced in the appendix. It turns out that different data should be clustered bydifferent and special cluster methods.