DETERMINING THE NUMBER OF CLUSTERS FOR MARKETING BINARY DATA.

In the article a new way of determining the number of clusters was proposed focused on data made up of binary variables. An important application aspect is that the data sets on which the new formula was investigated were generated in the way characteristic for the marketing data following the work of Dimitriadou et al. [2002]. The new formula is a modification of the Ratkowsky-Lance index and proved to be better in some respects than this index, which was the best in the mentioned research. The modification proposed is based on measuring the quality of grouping into the predicted number of clusters and running the same index on the twice smaller set of objects comprising dense regions of the original data set.