Research Of Spatial Outliers Detection Algorithm

With the extensive application of the sensor technology, such as Satellite, CT imaging, the number and complexity of spatial data are growing fast, but the processing technology for traditional data is relatively backward. Therefore, spatial data mining has become a new area of data mining research. Spatial outlier mining is an important branch of spatial data mining, it is used to find a spatially referenced object whose non-spatial attribute values are significantly different from the values of its neighborhood, and it can give important meaning in the applications of traffic control, sensed image analysis and others.On the basis of the comprehensive analysis for present research situation at home and abroad, further deep research is done for the algorithms of spatial outlier mining technology.Firstly, since existing spatial outlier detection algorithms have some disadvantages, such as a spatial object will be impacted by its spatial neighbors which have exceptionally large or small non-spatial attribute values, so in this book, based on the principle of K-Nearest Neighbor Graph and used the cutting edge strategy, we have proposed a KNN Graph-based algorithm to identify spatial outliers. The proposed algorithm can identify accurate spatial outliers and detect region outliers at the same time.Secondly, because existing spatial outliers detection algorithms mainly ignore the effects of spatial attributes which have more effects on the calculation of outlier degree. In order to improve the accuracy of spatial outliers mining, based on the ideology of Weighted Spatial Outlier, we use the method of weighted non-spatial attribute values and make it as the factors in the calculation of outlier degree. On the basis of the traditional Z-value algorithm, we propose an improved algorithms---Improved Z-value Algorithm.Thirdly, since the Improved Z-value algorithm exists the problem which can hide the difference between spatial object and its neighbors when we compute the outlier degree of spatial object, we propose the Weighted Difference Algorithm---WDA based on the ideology of Weighted Spatial Outlier and the Improved Z-value algorithm.Finally, all the algorithm for spatial outliers mining based on the KNN Graph and the improved algorithms based on the ideology of WSO are implemented on the FMR datasets and WNV datasets to validate the effectiveness of the proposed algorithms.