A study on spatial data mining using Geo-CBR and its application

Currently,Geo-data-mining and knowledge discovering,a new kernel of GIS spatial analysis study,which help to break theoretic limitation of Geo-expert system and to reveal an innovative research roadmap for new era Geo-information sciences,represent latest trend in researching GIS.Various research communities have tried to apply or revise mathematic tools as probability theory,spatial statistic,fuzzy set and rule based induction method to studies concerning specific geo-scientific problems.According to the latest decade development in this study area,data mining method has absorbed,borrowed and revised latest mathematic tools and theories rising in AI study area;and focused both on theoretic research and its application in mining rules lying in spatial dataset.Development of Geo-data-mining couples tightly with AI and application mathematics by widely crossing and deeply fusing. CBR(Case Based Reasoning),a new AI method that expands knowledge capturing channels,encapsulating problems by case,solving new problem by referencing historical similar ones,storing and re-using successful cases,has advantages such as simplicity,flexibility,scalability,high efficiency,knowledge learning and accumulation,which enable CBR to analyse and reason complex geo-problems. This paper mainly discusses Geo-CBR from a spatial data mining view and deems it as a kind of problem oriented spatial data mining method.Firstly,a detailed Geo-CBR definition and its encapsulating method are given as well as discrimination between spatial data mining and problem oriented Geo-CBR.Then,considering physical geography zonal and regional variation effect,inter-dependent and mutually condition relationships between geo-cases are examined in depth.And a quantitative data-mining method to explore intrinsic spatial relationships from geo-cases is presented based on rough set theory.In addition,due to variation of spatial feature types and their spatial relationships in geo-case representative model,3 categories of spatial similarity calculating models are derived.Finally,a pilot study for LU is provided with purposes of landuse problems quantitative analysis and deduction and demonstration of Geo-CBR's characteristics and advantages in solving and analysis spatial related problems.