Cluster and visualize data using 3D self-organizing maps

The Self Organizing Maps (SOM) can be considered as an excellent computational tool and has been applied in numerous application areas. The SOM can be effectively used to visualize and explore the properties of multidimensional data. In this paper, the structure of traditional SOM map has been extended to a three-dimensional Self Organizing Maps (3D-SOM) maps. The purpose of this work was to study the ability of SOM´s algorithm and structure for data clustering in 3D space. Extensions of SOM algorithm in terms of the number, relation and structure arrangement of its output neurons, neighbourhood weight update processes and distance calculations in 3D space has been developed. The proposed method has been demonstrated by computing it on Iris flowers dataset using high level computer language. The performance of traditional SOM and 3D-SOM in terms of quantization and topographic errors has been compared and discussed. The results have shown that 3D-SOM has been able to cluster data and form a 3D data representation. The proposed technique has also reduced both quantization error (17.76%) and topographic error (32.93%) compared to usual SOM technique.