Chameleon: Hierarchical Clustering Using Dynamic Modeling

68 Computer C lustering is a discovery process in data mining. 1 It groups a set of data in a way that maximizes the similarity within clusters and minimizes the similarity between two different clusters. 1,2 These discovered clusters can help explain the characteristics of the underlying data distribution and serve as the foundation for other data mining and analysis techniques. Clustering is useful in characterizing customer groups based on purchasing patterns, categorizing Web documents, 3 grouping genes and proteins that have similar func-tionality, 4 grouping spatial locations prone to earthquakes based on seismological data, and so on. Most existing clustering algorithms find clusters that fit some static model. Although effective in some cases, these algorithms can break down—that is, cluster the data incorrectly—if the user doesn't select appropriate static-model parameters. Or sometimes the model cannot adequately capture the clusters' characteristics. Most of these algorithms break down when the data contains clusters of diverse shapes, densities , and sizes. Existing algorithms use a static model of the clusters and do not use information about the nature of individual clusters as they are merged. Furthermore, one set of schemes (the CURE algorithm and related schemes) ignores the information about the aggregate interconnectivity of items in two clusters. The other set of schemes (the Rock algorithm, group averaging method, and related schemes) ignores information about the closeness of two clusters as defined by the similarity of the closest items across two clusters. (For more information, see the " Limitations of Traditional Clustering Algorithms " sidebar.) By only considering either interconnectivity or close-ness, these algorithms can easily select and merge the wrong pair of clusters. For instance, an algorithm that focuses only on the closeness of two clusters will incorrectly merge the clusters in Figure 1a over those in Figure 1b. Similarly, an algorithm that focuses only on interconnectivity will, in Figure 2, incorrectly merge the dark-blue with the red cluster rather than the green one. Here, we assume that the aggregate interconnec-tivity between the items in the dark-blue and red clusters is greater than that of the dark-blue and green clusters. However, the border points of the dark-blue cluster are much closer to those of the green cluster than those of the red cluster. Chameleon is a new agglomerative hierarchical clustering algorithm that overcomes the limitations of existing clustering algorithms. Figure 3 (on page 70) provides an overview of the overall approach …