Research On Cluster Analysis For Spatial Data Mining

As one of the most important techniques and methods in Spatial Data Mining, Spatial Clustering Analysis has received wide attention. Different from classification, clustering aims at making elements of the same cluster more similar and widening the gaps among elements of different clusters, without any prior knowledge, and the clusters are made by the similarity of data. Being a module in Spatial Data Mining System, Spatial Clustering Analysis not only can identify clustering principles buried in large amounts of spatial data, but also serves as a pretreatment step of other data mining algorithm in assisting other data mining methods to dig out more profound knowledge, so as to improve the overall efficiency and quality of data mining. Therefore, it is of great significance as to how to improve the performance of Spatial Clustering Algorithm, in order to improve its quality and efficiency and meet the needs of practical application in the objective world. This article focuses on the following points in researching on the Spatial Clustering Analysis methods:(1) It briefly introduces the basic theories of spatial clustering, accounts the process of clustering and summaries the five representative spatial clustering methods as well as their strengths and weaknesses in a systematical way.(2) Simulated Annealing has the feature of seeking the best solution. This article introduces the basic theories of Simulated Annealing, which are employed to optimize K-means algorithm, the traditional spatial clustering method. The optimized method drops the local minimum by means of accepting inferior solution with probability, thus providing the possibility to seeking overall optimal solutions. Besides, it proposes the idea of point density in the process of optimization, which makes the clustering results unaffected by the initial value and promotes its efficiency as well.(3) It proposes a spatial clustering method based on triangular network off-line, according to relevant features of triangulation. First, the initial division of spatial entity is obtained through constructing triangular network and off-line. Then, the final clustering result is reached through further clustering by K-means algorithm. This method is a modification to traditional clustering methods. It not only improves the quality of clustering results, but also excels in identifying clustering with random figure.(4) As to the spatial clustering problem of existing obstacle restriction in the real world, this article proposes a method of seeking obstacle distance by means of convex hull. It modifies traditional k-mediods algorithm using micro-clustering techniques and the method of defining distance array. Based on the obstacle distance seeked and the clustering made by modified k-mediods algorithm, the clustering results regardless of obstacle-restriction are received.