Algorithm for Fast Spatial Outlier Detection

Because spatial data are usually high-dimensional, complex and mass, we categorize the attributes of each spatial data object as spatial attributes and non-spatial attributes. We use spatial attributes to construct spatial index and determine spatial neighborhood, and use non-spatial attributes to compute outlying degree and spatial outlying degree factor, so as to solve the problem of index and the measurement of outlying degree. In addition, we propose two heuristic pruning strategies to realize fast pruning away those can not be candidate outliers in the data set. According to spatial self-correlation, the impact extent of neighborhood is added to compute attribute weighted values. At the same time, the weighted values are added to calculate pair-wise distance of each spatial object. In this paper, we propose a novel measure, spatial outlying degree factor (SODF), which captures the local behavior of datum in its spatial neighborhood. The experimental results show that the proposed SODF algorithm outperforms the other existing algorithms in detection accuracy, scalability, user dependency and efficiency.