Knowledge Discovery by Spatial Clustering Based on Self-organizing Feature Map and a Composite Distance Measure

This paper proposes a spatial clustering model based on self-organizing feature map and a composite distance measure, and studies the knowledge discovery from spatial database of spatial objects with non-spatial attributes. This paper gives the structure, algorithm of the spatial clustering model based on SOFM. We put forward a composite clustering statistic, which is calculated by both geographical coordinates and non-spatial attributes of spatial objects, and revising the learning algorithm of self-organizing clustering model. The clustering model is unsupervised learning and self-organizing, no need to pre-determine all clustering centers, having little man-made influence, and shows more intelligent. The composite distance based spatial clustering can lead to more objective results, indicating inherent domain knowledge and rules. Taking urban land price samples as a case, we perform domain knowledge discovery by the spatial clustering model. First, we implement several spatial clustering for multi-purpose, and find series spatial classifications of points with non-spatial attributes from multi-angle of view. Then we detect spatial outliers by the self-organizing clustering result based on composite distance statistic along with some spatial analysis technologies. We also get a meaningful and useful spatial homogeneous area partition of the study area. * Corresponding author: Liu Yaolin, E-mail:yaolin610@163.com; This paper is funded by 863 program of China (2007AA122200), 973 program of China (2006CB701303) and Open Research Fund Program of GIS Laboratory of Wuhan University(wd200609)