Most of unsupervised learning algorithms use a dissimilarity function to measures similarity
between the objects within the dataset. However, traditionally dissimilarity
functions did not design and fail to treat all spatial attributes of region or just solve
partial kinds of region since incomplete representation of structural of region and other
spatial information contained within the region datasets. In this research, we modified
polygonal dissimilarity function (PDF) that comprehensively integrates both the spatial
and the non-spatial attributes of a polygon to specifically consider the density and
distribution that exist within the region datasets and work well to regular region, but
not for irregular region. We represent a polygon as a set of intrinsic spatial attributes
by slice vertices and structural region, extrinsic spatial attributes, and non-spatial attributes.
Modified PDF was applied on cluster validity method by Davies Bouldin (DB)
and spatial clustering by using CLARANS. Spatial clustering by CLARANS with modified PDF using two characteristically different sets of data, (a) regular geometric shapes
(dummy region) and (b) irregular geometric shapes, Jakarta crime as case study on
spatial clustering. Modified PDF is working and does not has member of disjoint cluster
for more unstructured regions compared to origin PDF, has two percents smallest
SSE than PDF, and accurately 36 percent than PDF on spatial clustering. Completely
spatial information has above fifty percents significances and best cluster result for all
dataset. In addition, modified PDF with DB (DBP ) can evaluated result of spatial
region clustering than Silhouette index.