GROUPALS: A method to cluster objects with mixed measurement levels


This paper proposes a method to cluster objects that are measured on a set of categorical variables with mixed measurement levels. The new feature of the approach is that the scaling of variables, and thereby the construction of distances between objects, and the clustering of objects are performed simultaneously. The method works by minimizing the value of a loss function that finds principal components of the input variables and that has been constrained by a restriction on the object scores. A computer program, called GROUPALS, using an alternating least squares algorithm, was developed to apply the method to actual data.

RR-86-10, Department of Data Theory, University of Leiden