Multivariate mixture distribution model based cluster analysis

Mixture distribution model is used for model based cluster analysis. Clusteranalysis based on mixture distribution models, is called as model based clusteranalysis. The aim of the model based cluster analysis is minimizing the distancesbetween each elements in a group and maximizing the differences between groups inmultivariate data set. In model based cluster analysis approach, it is assumed that themultivariate data is generated by a mixture distribution model in which eachcomponent corresponds to a different cluster in multivariate data set. Themultivariate probability density function for each component in mixture distributionmodel explains the structure of the corresponding cluster in multivariate data set. Inthis thesis: i) A new method is proposed for determining the number of componentclusters in the multivariate normal mixture model based cluster analysis by assumingthat the multivariate data having mixture distribution model and each component inmixture distribution model is multivariate normal distribution. ii) A new algorithm isdeveloped for multivariate normal mixture distribution model based cluster analysisby refining groups in multivariate data set. iii) The new algorithm developed formultivariate normal mixture distribution model based cluster analysis by refininggroups in multivariate data set is applied for supervised per field classification ofremotely sensed multispectral image data of an an agricultural region in Adana.