Robust classification in market research

Abstract In market research we usually face the problem of classifying multivariate data which show a high degree of correlation, are likely to contain several multivariate outliers and are not normal. In these situations the application of standard clustering methods like k-means which assume spherical data and absence of outliers fails completely to reveal the real structure of the data. In this work we introduce a dataset of facial cosmetics produced by industrial brand and private labels to illustrate the problems described above. We show, through an application, how it is possible using new robust classification tools and how is possible to cope with multivariate outliers, non normality and high correlation among variables.