Fast SNN-Based Clustering Approach for Large Geospatial Data Sets

Current positioning and sensing technologies enable the collection of very large spatio-temporal data sets. When analysing movement data, researchers often resort to clustering techniques to extract useful patterns from these data. Density-based clustering algorithms, although being very adequate to the analysis of this type of data, can be very inefficient when analysing huge amounts of data. The Shared Nearest Neighbour (SNN) algorithm presents low efficiency when dealing with high quantities of data due to its complexity evaluated in the worst case by O(n(2)). This chapter presents a clustering method, based on the SNN algorithm that significantly reduces the processing time by segmenting the spatial dimension of the data into a set of cells, and by minimizing the number of cells that have to be visited while searching for the k-nearest neighbours of each vector. The obtained results show an expressive reduction of the time needed to find the k-nearest neighbours and to compute the clusters, while producing results that are equal to those produced by the original SNN algorithm. Experimental results obtained with three different data sets (2D and 3D), one synthetic and two real, show that the proposed method enables the analysis of much larger data sets within reasonable amount of time.