The Research On Dimensionality Reduction Algorithm In Spatial Data Mining

Geographic information system is a computer system which supports to acquire,manage, process, analysis, model and display the spatial data. It is designed in order to solve the complicated problems of geography. As an important branch of data mining, spatial data mining could discover the implied and valuable information by data mining algorithm in the spatial database. It can make up their deficiencies and make them more powerful by combining with the geographic information system and the spatial data mining. By the continuous expansion of the spatial data attributes making a large number of redundant information, it is particularly important for this attribute compression now.This paper first introduces the concepts, characteristics and functions of the geographic information system and the spatial data mining. Then some representative examples of the dimensionality reduction algorithms in spatial data mining are presented in detail. On this basis, the paper carried out the following research:(1)Based on the orthogonal function system and FCM algorithm, it proposes a new method to clustering time series. First the time series with length n are first mapped in L2 space. Then the similarity of series will be obtained by computing the distance among functions. Last, the cluster results of time series have been achieved by FCM algorithm according to these similarities.(2) A dimensionality reduction technique to improve-Isomap method is proposed by weighted connections between neighborhoods. It attempts to preserve perfectly the relationships between neighborhoods in the process of dimensionality reduction.Based on the theories mentioned above, the validity of the proposed methods is tested by five typical examples which are widely employed for many clustering algorithms. The experimental results show that the desired cluster results can also be obtained under the case of 90% compressibility for given dataset. The experimental results obtained by improve-Isomap method show that the local topology nature of dataset is preserved well while transforming dataset in high-dimensional space into a new dataset in low-dimensionality by the proposed method. Finally, the clustering algorithm proposed in this dissertation is applied to meteorological data of Liaoning Province and the temporal and spatial distribution characteristics of meteorological factors are analyzed.