Miscellaneous Clustering Methods
The methods described in the preceding chapters form the major part of the body of work on cluster analysis. Nevertheless, there remain a substantial number of other methods that do not fall clearly into any of the previous categories, and in this chapter an attempt is made to describe a number of these techniques. While a comprehensive review is impossible simply because of the vastness of the literature involved, this situation is less daunting than it first appears, since some of the apparently specialized techniques are, in essence, very similar to standard clustering techniques. For example, some of the techniques developed in genetic research entail applying hierarchical clustering (see Chapter 4) but using a specialized distance measure such as Jukes–Cantor or the optimal matching coefficient (see Chapter 3). A method patented by the Hewlett Packard Development Company for text clustering uses a recursive hierarchical technique, but considering parts of speech in order (nouns, then verbs, then adjectives, for example) – see Kettenring (2009). Similarly, some of the newer pattern recognition techniques in imaging are closely related to traditional methods. The Kohonen self-organizing map, for example, discussed in Section 8.8.2 as an example of a neural network, is similar in principle to the k-means clustering technique described in Chapter 5.
The methods to be discussed in this chapter can be categorized as follows: ...