Towards B-Coloring of SOM

Abstract

The Self-Organizing Map (SOM) is one of the most popular neural network methods. It is a powerful tool in visualization and analysis of high-dimensional data in various application domains such as Web analysis, information retrieval, and many other domains. The SOM maps the data on a low-dimensional grid which is generally followed by a clustering step of referent vectors (neurons or units). Different clustering approaches of SOM are considered in the literature. In particular, the use of hierarchical clustering and traditional k-means clustering are investigated. However, these approaches don't consider the topological organization provided by SOM. In this paper, we propose BcSOM, an extension of a recently proposed graph b-coloring clustering approach for clustering self organized map. It exhibits more important clustering features and enables to build a fine partition of referents by incorporating the neighborhood relations provided by SOM. The proposed approach is evaluated against benchmark data sets and its effectiveness is confirmed.