Spatial Data Mining Classification Method And Its Application

With the application of spatial technology in every industry and the quick development of data mining, spatial data collection technology, database technology, the demand of discovering knowledge from spatial database strengthens increasingly and a new research field in order to get knowledge from spatial database has appeared- spatial data mining. Spatial data mining is the process of distill the connotative spatial, non-spatial pattern and common character that the people are interested in from spatial database, it is a new area that integrates multi-subject and many technologies and combines the technologies of machine learning, spatial database system, expert system, mobile compute, statistics, remote sensing, system based on knowledge, visualization and so on. Though at first data mining comes from relation database and transactional database, with the special character of spatial data, mining knowledge from spatial database brings the attention of data mining researchers.This paper introduces the basic theory of spatial data mining systemically, compares the differences between the traditional data mining and spatial data mining, and then analyses the research actuality and the current problems of data classification in spatial data mining. Through the study of existed methods and combination of statistics, rough set theory, information theory and son on, solutions of continuous data's discretization, predication selection and data classification in spatial data mining are submitted and a spatial data mining archetypal system is achieved.The main work of this paper is listed as follows: Based on the analysis of variance in statistics and the idea that the total square sum of data can be separated into efficacy square sum and error square sum, this paper submits a continuous data discretization algorithm based on the maximum efficacy square sum, we initialize the data by histogram method, then compute the relevant efficacy square sum and error square sum until the efficacy square sum is maximal. This paper analyzes the relationship of different level spatial objects throughthe maximum entropy principle; pick out the spatial predication that can fully incarnate the relationship of spatial objects and relevant non-spatial dataset through mutual information, Z-test and Kullback-Leibler distance as an iterative condition.? A basic aspect that needs uncertain disposal in spatial data mining is topological relation of spatial objects. This paper gives a spatial data classification method that deals with spatial relationship based on probability rough set model, this method can handle with uncertain relationship of spatial data, reduce the data with keeping information classification purpose unchanged, manage many kinds of noisy and output the spatial data classification rules finally.? A spatial data mining archetypal system is achieved by the research of data preprocessing, spatial prediction selection and data classification in spatial data mining. This system is built under windows OS, Visual C++ 6.0 and MapX5.0, which achieves data preprocessing, spatial feature selection, spatial data classification, rules selection function and so on. It has a visual interface, accomplishes basic GIS operation and layer management and so on, so we can manipulates spatial data easily.