Knowledge Discovery in Spatial Planning Data: A Concept for Cluster Understanding - Computational Approaches for Urban Environments - page 59

1

0.5

0

−0.5

−1

−1

0

2

1.5

x

1

0.5

0

1 −0.5

−1

y

Fig. 3.10 Chain-link dataset

objects. A valley describes similar objects, characterized by small U-heights on the
U*-Map. Data points found in coherent regions are assigned to one structure. All
local regions lying in the same structure have nearly the same properties. Outlier
groups can also be clearly discerned, characterized by small valleys and very
prominent boundaries (large U-heights). Close inspection of the U-matrix reveals
that two “structures” can in fact be regarded as outlier, having only two members
each. The first group of outlier consists of the major cities Munich and Berlin, both
of which display a large value for sealed surfaces, a small value for building area per
inhabitant, and low value for protected areas. The second group of outlier consists
of Suhl and Baden-Baden, two cities which are remarkable for their extraordinarily
large proportion of protected areas and low value for sealed surfaces.

3.3.6 Data Clustering

After an initial inspection of the multidimensional data space, the next important
step is to undertake a partitioning into coherent data structures. Groups must be
formed of similar data objects which can be easily distinguished from the remaining
data. This task is termed clustering. Depending on the type of resulting groups,
cluster processes can be subdivided into hierarchical, partitional, or overlapping
clustering (Hand et al. 2001 ).