A novel dual-domain clustering algorithm for inhomogeneous spatial point event. (26th October 2020)

• Record Type: Journal Article
• Authors: Zhu, Jie; Yang, Jing; Di, Shaoning; Zheng, Jiazhu; Zhang, Leying
• Is Part Of: Data technologies and applications. Volume 54:Number 5(2020)
• Page Start: 603
• Page End: 623
• Publication Date: 2020-10-26
• Subjects: Dual clustering -- Delaunay triangulation -- Inhomogeneity -- Spatial proximity -- Attribute similarity -- Data mining; Information science -- Periodicals; Electronic information resources -- Periodicals; Knowledge management -- Periodicals; 020.5
• DOI: 10.1108/DTA-08-2019-0142
• Languages: English
• ISSNs: 2514-9288

Abstract

Abstract : Purpose: The spatial and non-spatial attributes are the two important characteristics of a spatial point, which belong to the two different attribute domains in many Geographic Information Systems applications. The dual clustering algorithms take into account both spatial and non-spatial attributes, where a cluster has not only high proximity in spatial domain but also high similarity in non-spatial domain. In a geographical dataset, traditional dual spatial clustering algorithms discover homogeneous spatially adjacent clusters suffering from the between-cluster inhomogeneity where those spatial points are described in non-spatial domain. To overcome this limitation, a novel dual-domain clustering algorithm (DDCA) is proposed by considering both spatial proximity and attribute similarity with the presence of inhomogeneity. Design/methodology/approach: In this algorithm, Delaunay triangulation with edge length constraints is first employed to construct spatial proximity relationships amongst objects. Then, a clustering strategy based on statistical change detection is designed to obtain clusters with similar attributes. Findings: The effectiveness and practicability of the proposed algorithm are illustrated by experiments on both simulated datasets and real spatial events. It is found that the proposed algorithm can adaptively and accurately detect clusters with spatial proximity and similar non-spatial attributes under the consideration of inhomogeneity. Originality/value: Traditional dual spatial clustering algorithms discover homogeneous spatially adjacent clusters suffering from the between-cluster inhomogeneity where those spatial points are described in non-spatial domain. The research here is a contribution to developing a dual spatial clustering method considering both spatial proximity and attribute similarity with the presence of inhomogeneity. The detection of these clusters is useful to understand the local patterns of geographical phenomena, such as land use classification, spatial patterns research and big geo-data analysis.