Robust Outlier Detection Method For Multivariate Spatial Data

Spatial data consist of spatial attributes describing the topology of the object and non-spatial attributes carrying information on the behavioral aspects of the object. This object is termed as a spatial outlier if its non-spatial attributes are significantly different from those in its spatial neighborhood. Here, a robust algorithm based on the Comedian approach is proposed for the detection of spatial outliers in multivariate spatial data. A simulation study is carried for comparisons of the proposed procedure with an existing MCD based spatial outlier detection technique. The simulation results show that the proposed algorithm outperforms the existing algorithm. For demonstrating the effectiveness of the proposed algorithm, its application on real-life CRIME data of India is discussed.