Phillip Bermingham

Cluster analysis has become part of the spatial analyst's tool set, and has enabled and enhanced the analysis of multivariate spatially distributed phenomena in a variety of applications. However, the spatial proximity of observations is generally not considered in the allocation process. The purpose of this research was to identify a methodology that would allow spatial proximity to be combined with the more traditional attribute similarity in a multivariate clustering procedure. In doing so; the combined dimensions ideally would be able to identify groups within the data that are both homogeneous and spatially compact. In addition, it would be preferable for the methodology to be flexible so that the influence of spatially proximity on the solution could be easily altered. This is accomplished through a multiplicative weighted combination of spatial proximity and traditiqnal attribute similarity. Besides developing the methodology, a working algorithm was also developed to demonstrate the model using real datasets.