NEC: A hierarchical agglomerative clustering based on fisher and negentropy information

Abstract

In this paper a hierarchical agglomerative clustering is introduced. A hierarchy of two unsupervised clustering algorithms is considered. The first algorithm is based on a competitive Neural Network or on a Probabilistic Principal Surfaces approach and the second one on an agglomerative clustering based on both Fisher and Negentropy information. Different definitions of Negentropy information are used and some tests on complex synthetic data are presented. © Springer-Verlag Berlin Heidelberg 2006.