kd-SNN: A Metric Data Structure Seconding the Clustering of Spatial Data

Large amounts of spatio-temporal data are continuously collected through the use of location devices or sensor technologies. One of the techniques usually used to obtain a first insight on data is clustering. The Shared Nearest Neighbour (SNN) is a clustering algorithm that finds clusters with different densities, shapes and sizes, and also identifies noise in data, making it a good candidate to deal with spatial data. However, its time complexity is, in the worst case, O(n(2)), compromising its scalability. This paper presents the use of a metric data structure, the kd-Tree, to index spatial data and support the SNN in querying for the k-nearest neighbours, improving the time complexity in the average case of the algorithm, when dealing with low dimensional data, to at most O(n x log n). The proposed algorithm, the kd-SNN, was evaluated in terms of performance, showing huge improvements over existing approaches, allowing the identification of the main traffic routes by completely clustering a maritime data set.