Synthesized clustering: A method for amalgamating alternative clustering bases with differential weighting of variables (Q1077114)

A new clustering method, called SYNCLUS (SYNthesized CLUStering) is proposed for dealing with the problems of weighting the variables in a specific battery and of weighting alternative sets of variables as fragments of the whole battery (e.g. demographics, psychographics, product usage, etc. in marketing research). The model for this method is presented and a detailed description of the algorithm is given, being composed of seven phases (input and preprocessing; distances calculation; generalized K-means; definition of cluster distances; solution for the importance weights; test for convergence; output). SYNCLUS Monte Carlo results on synthetic data are compared to results for ordinary K-means and average-link hierarchical clustering. Comparison to the known clustering methods is also given when applied to a set of real data, obtained from a marketing research study of physician's attitudes toward a new antihypertensive drug (media judgments, ''life style'', and demographics as variable subsets). A number of possible applications of SYNCLUS are suggested such as marketing research and segmentation, behavioral science, educational psychology. Possible algorithmic enhancement and method generalizations are discussed, and limitations of SYNCLUS are indicated.