The GIS revolution and the increasing availability of GIS databases emphasize the
need to better understand the typically large amounts of spatial data. Clustering is
a fundamental task in Spatial Data Mining and many contributions from researchers
in the field of Knowledge Discovery are proposing solutions for class identification
in spatial databases. The term spatial data refers to a collection of (similar) spatial
objects, e.g. areas, lines or points. In addition to geographic information, each
object also possesses non-spatial attributes. In order to apply traditional data mining
algorithms to such data, the spatial structure ans relational properties must be made
explicite. SCOT deals with the special case of grouping German towns. The towns
are related to each other by the various streets connecting them. Each town also
possesses an inner spatial structure, the local street network, and further non-spatial
information. This thesis considers all three kinds of information for the clustering
of towns. It exploits the concept of neighborhood to capture relational constraints,
measures the similarity of the structures of local street networks and transforms the
most important non-spatial attributes. SCOT is part of a project at Fraunhofer IAIS,
Germany, and has been successfully applied in practice.