An efficient topological-based clustering method on spatial data in network space

In recent years, along with the rapid development of location-based information, clustering algorithms on spatial data have been extensively applied for data insight and knowledge discovery. Existing clustering techniques usually depend on user parameters and mainly perform on the plane with Euclidean distance. This work proposes a Network Space Topological-Based Clustering (NS-TBC) algorithm for clustering network-constraint objects using a topological-based framework. This approach replaces the distance measures with topological relations for the spatial clustering problem in network space to optimize the parameters. The proposed method exploits the advantages of the ACUTE algorithm, proposed in 2016, but for network-constraint objects. The NS-TBC algorithm is applied to six datasets from Open Street Map to demonstrate its effectiveness. It outperforms the internal validation Davies-Bouldin index against the ACUTE and iNS-DBSCAN algorithms, in which iNS-DBSCAN is our state-of-art efficient clustering algorithm for network-constrained spatial data. The runtime is also carefully measured for evaluation purposes. The evaluation was performed only for two algorithms, NS-TBC, and iNSDBSCAN, because ACUTE was not designed to work in network space. The experimental results show that the NS-TBC algorithm uses less than 50% of the computation time needed by the iNS-DBSCAN algorithm. In short, the proposed algorithm NS-TBC provided a solution to reduce the number of parameters for the iNS-DBSCAN algorithm while significantly improving the execution time and cluster quality.