A New Method of Spatial Outlier Detection by Considering Multi-level Thematic Attribute Constraints

Spatial outlier detectionis an important approach in spatial data mining and knowledge discovery. Spatial outliers are entities whose non-spatial attributes are significantly different from the value of other entities in their spatial neighborhoods. The current methods have limitations in capturing spatial neighborhoods and detecting small abnormal clusters. In order to solve this problem, we develops a new method of spatial outlier detection that considers thematic attributes, named MTACSOD. Firstly, a constrained Delaunay triangulation is used to construct reasonable and stable spatial proximity relationship. Then, for the thematic attribute distance between adjacent spatial entities, global and local constraints are imposed consecutively to refine spatial adjacency. Finally, a spatial outlier identification index is proposed to detect spatial outliers. Both simulated and real-life datasets are used to illustrate the advance and practicability of the MTACSOD proposed in this paper.