Evaluation of Topographic Clustering and Its Kernelization

We consider the topographic clustering task and focus on the problem of its evaluation, which enables to perform model selection: topographic clustering algorithms, from the original Self Organizing Map to its extension based on kernel (STMK), can be viewed in the unified framework of constrained clustering. Exploiting this point of view, we discuss existing quality measures and we propose a new criterion based on an F-measure, which combines a compacity with an organization criteria and extend it to their kernel-based version.

• Type: Conference
• Year: 2003
• Where: ECML
• Authors: Marie-Jeanne Lesot, Florence d'Alché-Buc, George Siolas
• Keywords: ECML 2003, Topographic Clustering, Topographic Clustering Algorithms, Topographic Clustering Task