Efficient Voronoi K-Means Algorithm for Mining Local Crime Spatial Outliers in Spatial Crime Data

Through the boosting accessibility of spatial and temporal data in many research fields, spatial clustering and spatial outlier detection has received a group of concentration in the spatial data mining research. As a very famous method, the CLIQUE Optimization finds a region that deviates significantly from the entire spatial data set. In this paper, we introduce the novel problem of mining crime regional outliers in spatial crime data. A crime regional outlier is a rectangular area which contains an outlying object such that the difference between the sparse area and dense area, the value of this object and the gathered value of this crime attribute over all objects in the area is maximized. It compared to the CLIQUE Optimization, which targets crime global outliers, our research aims at the local crime spatial outliers. This research introduces Voronoi based K-Means algorithm for mining crime regional outlets, increasing areas by extending them by at least one neighbouring object per iteration, preferring the expansion which leads to the major increase of the objective function. The experimental evaluation on real datasets, the result shows the new type of crime regional diagram and the efficiency of the proposed algorithm.