Statistical Approach to the Identification of Separation Surface for Spatial Data

In spatial clustering, spatial objects are grouped into clusters according to their similarities. In terms of learning or pattern recognition, it belongs to the identification of structures/classes through an unsupervised process. In terms of data mining, it is the discovery of intrinsic classes, particularly new classes, in spatial data. It formulates class structures and determines the number of classes. I have examined in Chap. 2 the importance of clustering as a means for unraveling interesting, useful and natural patterns in spatial data. The process generally does not involve how to separate predetermined classes, or how to determine whether classes are significantly different from each other, or how to assign new objects to given classes. Another fundamental issue of spatial knowledge discovery involves spatial classification. It essentially deals with the separation of pre-specified classes and the assignment of new spatial objects to these classes on the basis of some measurements (with respect to selected features) about them. In terms of learning or pattern recognition, it is actually a supervised learning process which searches for the decision surface separating appropriately various classes. In terms of data mining, it often involves the discovery of classification rules from the training/learning data set that can separate distinct/genuine classes of spatial objects and the assignment of new spatial objects to these labeled classes. Whether the pre-specified classes are significantly different is usually not the main concern in classification. It can be determined by procedures such as the analysis of variance in statistics.