A Relative Approach to Hierarchical Clustering

This paper presents a new approach to agglomerative hierarchical clustering. Classical hierarchical clustering algorithms are based on metrics which only consider the absolute distance between two clusters, merging the pair of clusters with highest absolute similarity. We propose a relative dissimilarity measure, which considers not only the distance between a pair of clusters, but also how distant are they from the rest of the clusters. It defines for each cluster its relative nearest cluster respect to the whole data set, instead of its usual absolute nearest cluster. We have named the agglomerative hierarchical scheme with this relative dissimilarity measure as Relative Hierarchical Clustering (RHC). The performance of RHC is compared with the classical approach with different intercluster distances by using three data sets.