A multivariate spatial outlier detection method based on semantic similarity

Abstract

Spatial outlier detection is usually done in local neighbourhoods. We use an interpretable similarity measure to calculate the similarity between the regions of influence of spatial objects using spatial and non spatial attributes and relations both in the semantic description and in the similarity measure. A systematic method for similarity threshold selection is presented that can be used to categorize objects by their behavioral pattern in semantically similar neighborhoods. This paper's main contribution is developing a multivariate two level outlier detection method. Real world data is used to evaluate our method. © 2012 Springer Science+Business Media B.V.