A New Kind Of Two-Group Clustering And Discriminating Analysis Method And It's Empirical Studies

when carrying out clustering analyzing on a set of samples using classical clustering analysis methods, it sometimes gets different clustering results if using different variable or different combinations of variables. The reason is that different variable or different combinations of variables reflect different aspects of the samples. So it comes that variables have different influence on the result of clustering analysis. But classical clustering methods have not presented this difference in the result, and this is not good for better clustering results or better explanation for the clustering results.Another problem of the classical clustering analysis method is that, when the combination of variables used to carry out the analysis is set, usually different results are get when using different clustering methods. Generally, the commonness of the results is drawn out, and the disputable samples are left to the discriminating methods to decide their group. And when carrying out the discriminating analysis using different methods, different results are got. This brings out the problem of establishing the standard to evaluate different clustering and discriminating methods, but there is still no suitable one till today.To solve the above problems, a new kind of two-group discriminating algorithm is advanced in this paper after reviewing the theory of classicalclustering and discriminating methods. Then two kinds of dynamic clustering algorithm are worked out. The entire above algorithm is implemented using a computer program. Relevant theoretic research and empirical study are also carried out:(1) The review of the classical clustering and discriminating methods comes out that, the essential of clustering analysis is to find out a kind of combination of the samples among the limited ones, which not only fulfils especial clustering definition, but is also good for explanation, and the essential of discriminating analysis is to establish the standard for classifying. This offers the basal thought for the design of new methods. The above analysis also points out some problems existing in those classical methods, and it is good for resolving those problems in the new design.(2) A new kind of two-group discriminating algorithm is advanced. It partly resolves the problems existing in the classical methods.(3) Two designs of the new dynamic clustering method are brought out, and the optimal problems in the implementation of them are also partly figured out.(4) Relevant empirical studies are carried out, including contrasting analysis with the classical methods, and the results are acceptable.