A New Approach to Hierarchical Clustering Using Partial Least Squares

We here propose a methodology to improve hierarchical cluster analysis using partial least squares (PLS). Two problems are addressed by this methodology, these are (1) when, as usually, Euclidean distance is used for hierarchical cluster analysis, but Euclidean distance is defined only in Euclidean space. If Euclidean distances are computed in other spaces, the distances are hard to make sense. On the other hand, since the variables in the data set do not have equal variance, they do not have comparable scales. (2) Traditional clustering methods are based on single data table, but the application of PLS makes it possible to deal with multiply data tables problems. In addition, the proposed method can reduce the dimension of classification variables in a reasonable way. That makes it possible to demonstrate the relationship of multiply dimension data