Graphics Reference
In-Depth Information
ter five the correlation is much weaker. SPD and GRUENE are weakly correlated in
both clusters;thedifferencehereismainly the variance: GRUENEhasamuchhigher
variance in cluster four than in cluster five.
Whileall ofthesepatterns canofcoursealsobeinferredfromnumerical printouts
of the component parameters, they are made much more obvious by using graphical
techniques.Completescatterplotmatricesofallpairwisecombinationscanbeusedto
findinteresting variable combinations. Figures
.
and
.
could beaugmented by
aneighborhoodgraph similar tothose used inSect.
.
.Anatural choicetomeasure
the proximity of the mixture components is the Kullback-Leibler-divergence of the
component distributions, see Leisch (
) for details.
Summary
11.5
he natural visualization method for hierarchical clustering is the cluster dendro-
gram, because it directly reflects the construction principle of the underlying algo-
rithm.Similarly,thevisualization ofSOMSistightly bundledwiththealgorithm, and
feature maps on three-dimensional grids are a de facto standard.
Two larger groups of plots are available for partitioning and model-based clus-
tering: the first group are diagnostic plots like silhouettes and posterior rootograms,
which try to visualize the quality of the clustering. he second group of plots sim-
ply treats cluster membership as a categorical variable, and uses standard techniques
likeglyphs, colorsortrellisdisplaystohighlightclustermembershipinvisualizations
of the orginal data (or projections thereof). Virtually all of the methods of plotting
a group of variables against a single categorical variable proposed in this handbook
canbeusedforthispurpose.heexamples showninthischapterarepopularchoices,
but are primarily intended as a starting point and source of inspiration.
References
Becker, R., Cleveland, W. and Shyu, M.-J. (
). he visual design and control of
trellis display, Journal of Computational and Graphical Statistics
:
-
.
Everitt, B.S.,Landau, S.and Leese, M.(
). Cluster Analysis,
th edn, Arnold, Lon-
don, UK.
Fraley,C.andRatery,A.E.(
).Model-basedclustering,discriminantanalysis
and density estimation, Journal of the American Statistical Association
:
-
.
Friendly,M.(
).Visualizing Categorical Data, SAS Press, Cary, NC. ISBN
-
-
-
.
Gordon,A.D.(
).Classification,
nd edn, Chapman & Hall / CRC, Boca Raton,
FL, USA.
Hartigan, J.A. (
). Clustering Algorithms, Wiley, New York.
Hartigan, J.A. and Kleiner, B. (
). A mosaic of television ratings, he American
Statistician
(
):
-
.
Hartigan, J.A. and Wong,M.A. (
).Algorithm AS
:A k-means clustering algo-
rithm, Applied Statistics
(
):
-
.