Enhancement of spatial data analysis

Abstract

This thesis studies several problems related to clustering on spatial data. On spatial geographic data that distinguish between spatial attributes and normal attributes, it concentrates on how to efficiently incorporate spatial information into the learning of the mixture model. In detail, it explores data fusion in radial basis function network for regression and proposes a hybrid Expectation-Maximization approach to clustering. On general spatial data for which such a distinction is dropped, it continues to examine clustering from another two perspectives. At a higher level, it investigates consensus clustering that combines a given set of partitions into a consolidated one that is most compatible to them. As a complement to cluster analysis, outlier detection targets the finding of those rare and exceptional data that cannot be assigned to any cluster. A method is developed to find outliers whose local densities are very different from those of its neighbors.