Abstract
After classifying a dataset by k-means algorithm, whether or not each cluster should be split or be merged is evaluated by introducing a distortion-ratio map. The distortion-ratio map, which has two axes of the distortion-ratio on splitting and that on merging, is composed of partitional regions having the evaluation on splitting and merging, and each region provides a score for a cluster in the region. Measuring the goodness of clustering results through the map, we can estimate the appropriate number of clusters in the dataset, and acquire its corresponding clustering result.
