A new hierarchical-clustering combination scheme based on scatter matrices and nearest neighbor criterion

Abstract

In the field of pattern recognition, combination of different classifiers is a common method to improve classification accuracy. Recently, tendency to improve the function of clustering methods, specifically partitional clustering methods, is being increased. Generally, hierarchical clustering is preferred to partitional clustering when the number of exact clusters is undetermined or when we are interested in finding the relation between clusters. Most of the proposed methods for clustering combination are based on partitional clustering. In this paper, a new method for combination of hierarchical clustering is proposed. In this method, in the first step the primary dendrograms of base hierarchical clustering methods (such as Single Linkage and Complete Linkage) are converted to matrices. Then these matrices are synthesized together in a weighted procedure and led to a final description matrix. The weights are determined based on two criteria: clustering scatter matrices and nearest neighbour of each pattern. The results show improvement in function of combination method rather than base clustering methods.