Research on statistics data classification by introducing spatial distance into SOM

There are two main limitations in Principal Component Analysis (PCA) and other traditional statistical methods: firstly, these methods only examine similarity among objects based on their attribute data, but ignore the spatial relationship among them, which means regions of geographical proximity often have similar features that can not be completely described by only a few economic attributes. Secondly, the information included in data may be incomplete because of the shortcomings in collecting process, but the missing information must be complemented in some way for obtaining more accurate results. So in this paper, the authors proposed to introduce spatial distance into traditional statistical methods to overcome these two defects, and chose a neural network method for making improvements to analyze civilian-owned economic statistics data of Sichuan province in 2007, and finally worked out a series of classification results for different spatial distance weight values. Based on these classification results, the authors discussed the reasonable range of spatial distance factor, and interpreted the meanings of this improvement from the point of view of the economic geography. It is proved that the classification results are consistent with the actual situation in this research. How to determine the distance factor is the key of successful application of this method.