Minimum Spanning Tree Based Spatial Outlier Mining and Its Applications

Abstract

Spatial outliers are spatial objects whose nonspatial attributes are significantly different from the values of their neighborhoods. Detection of spatial outliers will provide the user with meaningful, interesting and potential information. Usually, algorithms for outlier mining on traditional business-oriented datasets are no longer applicable to spatial datasets. A new algorithm based on MST clustering is proposed in this paper to identify spatial outliers. The algorithm organically integrates the approach of minimum spanning trees and the density-based mechanism for outlier mining. Basic spatial structure characteristics of spatial objects are maintained by Delaunay Triangles and MST clustering is achieved by cutting off several most inconsistent edges. It turns out that the algorithm can find true spatial outliers, and it doesnt require any parameter for the algorithm be specified firstly. Experiments on real application problems indicate that the proposed algorithm is feasible and effective for identifying outliers from the large-scale spatial datasets.