CPCA-BASED ALGORITHM FOR ULTRAMETRIC CONSTRAINTS COMPARISON FROM CLUSTERING CRITERIA

We underline the main statistical methods that either use as input or perform as output a Gramian matrix that can be used in the Constrained Principal Component Analysis (CPCA). We consider an Euclidean distance matrix with an external system of explanatory variables. We design the eigen-analysis that allows to study the distances between the images of statistical units projected onto the subspace of the explanatory variables. We suggest an algorithm to point out the cluster criterion having the smallest distortion due to the ultrametric constraints.