An Approach to Clustering Models Estimation

There are numerous clustering algorithms and clustering quality criteria present at the moment. According to clustering tasks, different criteria can be used as a ground for choosing one or another clustering model. But choosing of clustering quality criteria is in turn not based on any formal approach and depends only on an analyst's intuition. We propose a systematic approach to clustering quality estimation by analyzing the structure of clustering models and extracting the factors of clustering quality. Further, we believe that clustering quality estimation requires estimation of knowledge or information produced in result of clustering. However, we show that existing information criteria can not be used as criteria for choosing clustering models and propose a new information criterion for clustering and examine it experimentally.