Hybrid Kohonen Self Organizing Map for the Uncertainty Involved in Overlapping Clusters Using Simulated Annealing

Abstract

The Kohonen self organizing map is widely used as a popular tool in the exploratory phase of data mining. The SOM (self organizing maps) maps high dimensional space into a 2-dimensional grid by placing similar elements close together, forming clusters. Recently research experiments presented that to capture the uncertainty involved in cluster analysis, it is not necessary to have crisp boundaries in some clustering operations. In this paper to overcome the uncertainty, an optimized clustering algorithm based on SOM which employs the rough set theory and the simulated annealing as a general technique for optimization problems is proposed. The optimized two-level stage SA-Rough SOM (simulated annealing - rough self organizing map) (first using SOM to produce the prototypes that are then clustered in the second stage based on the combination of rough set and simulated annealing) is found to perform well and more accurate compared with the crisp clustering methods (i.e. Incremental SOM) and reduces the errors.