A Study On Clustering Analysis Algorithms In Spatial Data Mining

With the development of data capture technologies and diversification of data capturing means, large amounts of geo-reference data has been collected. The collected data far exceed people's ability to analyze it. Thus, new and efficient data analytical tools are needed to discover knowledge from large spatial databases. The needs result in emergence of spatial data mining.Spatial clustering analysis is the main research field of spatial data mining. With the help of spatial clustering analytical tools, not only clustering rules can be extracted in a large column of spatial database, but when combining with other data mining methods knowledge hidden deeply can be discovered efficiently and effectively as well. It is main concern to design clustering algorithms for area geographical entities and raster spatial data. The contribution of this paper has been concluded as follows.l.Systemically analyzing and summarizing different spatial clustering algorithms that have been published in documents. The fitness, performance, advantages and disadvantages, and complexity of different algorithms have been compared in the paper.2. A distance computation method for simple polygons has been presented based on analyzing the properties of area geographical entities. The method has the following advantages: (1) adapting to distance computation between two simple polygons; and (2) getting the line segments (one chain from each polygon) of polygons when calculating distance. The line segments "face" each other in the sense that a vertex in either chain can "see" at least one vertex in the other chain and can be further applied to calculate geometry shape similarity between two simple polygons.3. A new partitioning clustering algorithm for area geographical entities based on genetic algorithms has been presented. Only depends on the fitness other than exterior information that the algorithm can search the best solution of clustering problem.4.Two clustering algorithms for area geographical entities based respectively on cluster division and cluster validity function have been designed aiming to solve the problem of clustering objects when the number of clustering is unknown. The performance of algorithms has been verified too.5. A criterion of shape similarity between line segments has been presented, which meets rotating invariance and translation invariance. Based on the criterion a clustering algorithm named CACSS has been designed. A criterion involving distance and geometry shape similarity has been presented too. A new version algorithm based on genetic algorithms and new version CLARANS have been rewritten respectively.6. Combining wavelet transformation and k-means algorithm, a new clustering algorithm for raster spatial data has been presented which can not only improve the efficiency but also assure the quality of clustering results.