Clustering with unconstrained hyperboxes
Fabio Massimo Frattale Mascioli, Antonello Rizzi, Massimo Panella, Giovanni Martinelli · 1999
In the present paper a new fuzzy clustering algorithm is presented. It is a modified version of the min-max technique. By relying on the principal component analysis, it overcomes some undesired properties of the original Simpson's algorithm. In particular, a local rotation matrix is introduced for each hyperbox according to the data subset of the related cluster, so that it is possible to arrange the hyperbox orientation along any direction of the data space. Consequently, the new algorithm yields more efficient networks, improving the match between the resulting clusters and local data structure.