Wine Tasting and a Novel Approach to Cluster Analysis

This paper proposes an alternate view for understanding clusters and for determination of the variables that play a significant role in cluster makeup. We explore the creation of a new categorical variable from a given set of variables by means of the output of a clustering algorithm. We postulate that this new variable can be seen as being comprised of a few “important” variables and explore how this new variable relates to the original variables. Sensitivity analysis and discriminant analysis are used to confirm the selection of important variables. Then we show that this method is able to identify those key variables which vary most significantly throughout the clusters.