Clustering based on distortion-ratio criterion

Clustering based on a distortion-ratio criterion for data sets with large statistical differences of class data is treated, where K-Means algorithm (KMA) cannot necessarily reveal the good performance. After obtaining cluster centers attaining the minimum squared-error distortion or its approximation by KMA, clustering based on the distortion-ratio criterion, whose criterion provides a measure of the validity of clustering, is introduced. When the criterion is not satisfied, a split and merge procedure based on the distortion-ratio measure is executed. Focusing on an interesting data set which is not resolved by KMA, clustering based on the distortion-ratio criterion and the split and merge procedure is investigated.