A Clustering Criterion Based on Distortion Ratios and Its Algorithms

Abstract

A clustering criterion based on distortion ratios and
its algorithms are proposed without offering the knowledge of
the number of clusters. Computing distortion ratios on splitting
and distortion ratios on merging for clusters of a data set,
the criterion function is defined as the mean of the Euclidian
distances between points of those distortion ratios and a reference
point. Three algorithms are proposed, whose algorithms are
designed to optimize the criterion function over the number
of clusters and partitions of the data set. Through several
classification experiments, the effectiveness of the criterion and
those algorithms is demonstrated.