Cluster validation overview Sample Clauses

Cluster validation overview. The term cluster validation is used to describe the process of evaluating the “goodness” of clusterings of data produced by cluster analysis or other class discovery methods (Xxxxx and Jain, 1979; Halkidi et al., 2001). Cluster validation can be measured using three types of criteria: (1) criteria that compare a clustering of the data to known labels of the data (i.e. known subtypes), (2) criteria that measure the “goodness” of a clustering without a comparison to known labels, and (3) criteria that compare two or more clusterings of the data to each other. Criteria that compare a clustering of the data to known labels are usually called external criteria (Halkidi et al., 2001; Handl et al., 2005; Xu and Xxxxxx, 2009). Monte Carlo analyses of cluster analysis techniques use external criteria to compare clusterings of the data given by clustering methods to the “known” clusterings (i.e. labels) of simulated data. External criteria can also be used to compare a clustering of data based on attributes of one type to a classification of the data based on other attributes, or to describe the agreement between two different classification systems. For example, a clustering of patients using genetic attributes can be compared to labels given by traditional histologic classification. In the context of class discovery, external criteria can be used to describe the agreement between two different partitions of the data. The term internal criteria is sometimes used to describe any criterion that measures the goodness of a clustering without comparison to external labels Some authors use the term in a more narrow sense, to describe methods that assign a statistical probability to a clustering of the data, and the term relative criteria to describe criteria meant to be used for comparison of several partitions of the data (e.g. Jain et al., 1999; Halkidi et al., 2001; Gan et al., 2007).