Data Clustering and Topology Preservation Using 3D Visualization of Self Organizing Maps

The Self Organizing Maps (SOM) is regarded as an excellent computational tool that can be used in data mining and data exploration processes. The SOM usually create a set of prototype vectors representing the data set and carries out a topology preserving projection from high-dimensional input space onto a low-dimensional grid such as two-dimensional (2D) regular grid or 2D map. The 2D-SOM technique can be effectively utilized to visualize and explore the properties of the data. This technique has been applied in numerous application areas such as in pattern recognition, robotics, bioinformatics and also life sciences including clustering complex gene expression patterns. In this paper, the structure of traditionally 2D-SOM map has been enhanced to a three-dimensional Self Organizing Maps (3D-SOM) maps. It has the purpose to directly cluster data into 3D-SOM space instead of 2D-SOM data clusters. The primary works mostly involved the extensions of SOM algorithm in particular the number, relation and structure arrangement of its output neurons, neighbourhood weight update processes and distances calculation in 3D xyz-axis. The proposed method has been demonstrated by computing 3D-SOM visualization on iris flowers dataset using high level computer language. The performance of 2D-SOM and 3D-SOM in terms of their quantization errors, topographic errors and computational time has been investigated and discussed. The experimental results have shown that the 3D-SOM has been able to form a 3D data representation, has slightly higher quantization error and computational time but performed better topology preservation than in 2D-SOM.