| Spatial data mining(SDM) from spatial database, which is a significant field of data mining, is the extraction of implicit knowledge, spatial relations and discovery of interesting characteristics and patterns which are not explicitly represented in spatial database.The importance of spatial data mining is more evident because of the great development of spatial database and large amounts of accumulated data.This thesis systematically discussed the basic theory of spatial data mining and spatial co-location patterns, and summed up the characteristics of classics mining algorithms.Then the paper make a litter improvement on the basis of previous algorithms.General studies on spatial association rules are base on conventional association rule algorithms, which treat spatial databases as usual data sets.co-location patterns algorithms meet the demand of mining spatial association rules effectively an exactly. Spatial co-location patterns represent the subsets of features whose instances are frequently located together in a geographic space.The co-location patterns mining algorithm based on KD-Tree introduce a new effective geometric spatial method, using KD-Tree to enumerate all instances which have a neighborhood relation, and then co-location patterns of size two are generated.Next, the algorithm exploits the apriori-gen to generate all k-size(k>3) co-location patterns.This method is more effective than the general method;Meanwhile, the paper raise a new co-location mining algorithm base on density due to the objects of a feature are usually distributed non-uniformly.A dynamic upper bound of the prevalence for a candidate is maintained, as a result, the overall number of joins to identify instance is reduced.In the last, the reslut of conducted experimentation show the correctness and the completeless of our approach.The result also illustrate the decreasement of the executive time on the performance.Meanwhile,the algorithms are realized based on component GIS. |
PDF Full Text Request