|
Abstract:
|
Spatial data has feature variables that can be used in statistical analysis, but it also contains additional location, or geographic, variables (usually latitude and longitude or some other x-y scale) that nonspatial data does not. Our approach uses both the feature and geographic variables in spatial data in a two-step process. First, the feature variables are clustered into groups. Second, the cluster results are projected onto a map using the geographic variables. The clusters are then refined using the geographic information so as to enhance spatial proximity, contiguity and other relevant geographic properties. Several variations on this basic approach are possible, including using results from the second step to recalculate the first step. We will present details and results of experiments with simulated and real data.
|
