Extension of DBSCAN in Online Clustering: An Approach Based on Three-Layer Granular Models

In big data analysis, conventional clustering algorithms have limitations to deal with nonlinear spatial datasets, e.g., low accuracy and high computation cost. Aiming at these problems, this paper proposed a new DBSCAN extension algorithm for online clustering, which consists of three layers, considering DBSCAN, granular computing (GrC), and fuzzy rule-based modeling. Firstly, making use of DBSCAN algorithms' advantages at extracting structural information, spatial data are clustered via DBSCAN into structural clusters, which are subsequently described by structural information granules (IG) via GrC. Secondly, based on the structural IGs, a series of granular models are constructed in the medium space, and utilized to form fuzzy rules to guide clustering on spatial data. Finally, with the help of structural IGs and granular rules, a rule-based modeling method is constructed in the output space for online clustering. Experiments on a synthetic toy dataset and a typical spatial dataset are implemented in this paper. Numerical results validate the feasibility to the proposed method in online spatial data clustering. Moreover, comparative studies with conventional methods and existing DBSCAN variants demonstrate the superiorities of the proposed method, as well as accuracy improvement and computation overhead reduction.

Automated summary

This paper proposes a new DBSCAN extension algorithm for online clustering, which combines DBSCAN, granular computing, and fuzzy rule-based modeling. The algorithm overcomes the limitations of traditional clustering algorithms in dealing with nonlinear spatial datasets, and shows superior performance in terms of accuracy and computation overhead reduction compared to conventional methods and existing DBSCAN variants.