A New Automated Hierarchical Clustering Algorithm Based on Emergent Self Organizing Maps

Abstract

Emergent Self Organizing Map (ESOM) has been shown as a powerful nonlinear data transformation and visualization method. In [13] based on ESOM and some of its derivatives, U-Matrix and P-Matrix, a powerful automated clustering algorithm, U*C, is introduced, and it is shown that the algorithm performs better than the some of the benchmark algorithms. However, the mentioned algorithm is suitable for partitional clustering tasks, while in most of the real cases, because of the nature of the data sets (not the ESOM training algorithm) a hierarchical structure in the data can be assumed. In this paper, based on the main ideas of U*C algorithm and underlying meaning of the U-Matrix, we introduced an automated hierarchical clustering algorithm, which performs well for real data sets. After testing with some test cases, we applied the proposed algorithm on a real data set, including different energy, ICT and Urban related indicators of European and central Asian countries. The proposed algorithm identified the hierarchical groups among the selected countries.