Preface to First Edition - lib

326 CLUSTER ANALYSISR> pottery_dist library("lattice")R> levelplot(as.matrix(pottery_dist), xlab = "Pot Number",+ ylab = "Pot Number")Pot Number4544434241403938373635343332313029282726252423222120191817161514131211109876543211210864201 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930313233343536373839404142434445Pot NumberFigure 18.4Image plot of the dissimilarity matrix of the pottery data.Mächler, 2003). The logarithms of the mass, period and eccentricity measurementsare shown in a scatterplot in Figure 18.6. The diagram gives no clearindication of distinct clusters in the data but nevertheless we shall continueto investigate this possibility by applying k-means clustering with the kmeansfunction in R. In essence this method finds a partition of the observationsfor a particular number of clusters by minimising the total within-group sumof squares over all variables. Deciding on the ‘optimal’ number of groups isoften difficult and there is no method that can be recommended in all circumstances(see Everitt et al., 2001). An informal approach to the numberof groups problem is to plot the within-group sum of squares for each par-© 2010 by Taylor and Francis Group, LLC