We propose two algorithms based on a new hierarchical partitioning around medoids clustering method originally developed for gene expression data. We are concerned with a different application; therefore, the dissimilarity between the objects has to be different and must be designed to deal with anthropometric features. Furthermore, one of the algorithms incorporates a different rule to split the clusters, which, in our case, provides better results. Our procedures not only make it possible to obtain optimal prototypes, but also to detect outliers. These outliers should be removed before defining prototypes so that the companies' market share can be optimized.
All the analyses are performed using the anthropometric database obtained from a survey of the Spanish female population.