Cluster analysis

In discriminant analysis, an optimal allocation rule between different groups is estimated from a training sample. The type and number of groups are known. In some situations, however, it is neither known whether the data can be divided into homogeneous subgroups nor how many subgroups there may be. How to find such clusters in previously ungrouped data is the purpose of cluster analysis. In music, one may for instance be interested in how far compositions or performances can be grouped into clusters representing different “styles”. In this chapter, a brief introduction to basic principles of statistical cluster analysis is given. For an extended account of cluster analysis see e.g. Jardine and Sibson (1971), Anderberg (1973), Hartigan (1978), Mardia et al. (1979), Seber (1984), Blashfield et al. (1985), Hand (1986), Fukunaga (1990), Arabie et al. (1996), Gordon (1999), Ho¨ppner et al. (1999), Everitt et al. (2001), Jajuga et al. (2002), Webb (2002).