David Lahoz, Fermín Mallor Giménez, Pedro Mateo
In this work we consider a clustering problem, that is, grouping a large set of elements described by a large set of non-negative variables into homogeneous categories that are not predefined. The novelty is that besides the quantitative criterion to differentiate variables, as usual, there is also a qualitative criterion, that takes into account whether or not the variables take the value zero . This problem was presented and studied by means of a certain family of parametric transformations of the data set and also by means of Multiple Factor Analysis in [1]. In our work, we consider these parametric transformations of the data set and we use Self-Organizing Maps ([4]) as clustering technique. We compare the results with those obtained in [1]. This problem came up when the managers of a big telecommunications company wanted to know about the structure of their clients¿ portfolio identifying different consumption profiles. The proposed methods have been tested on artificial data sets with similar statistical properties to those found in the real one
© 2001-2024 Fundación Dialnet · Todos los derechos reservados