Research On Spatial Clustering Based On Multiple Density-Ordered Trees And Field-theory

Spatial clustering is one of the most significant branches of spatial data mining.According to the complexity and peculiarity characteristics of spatial data,spatial clustering aims to discover potential and meaningful knowledge hidden in the intensely massive numbers of spatial objects.In this way,some difficult issues for traditional clustering can be effectively solved and the natural spatial agglomeration patterns from spatial objects can be completely revealed.Spatial clustering has been widely applied in various fields of real life,such as climate change,disease surveillance and so on.Most existing spatial clustering algorithms are limited by complicated spatial patterns and the problem of how to accommodate geometrical properties and attributes in spatial clustering.Thus,the methodology of adaptive spatial clustering analysis is crucial for our research.In order to overcome such limitations,we propose a novel clustering method based on Multiple Density-Ordered Trees and a modified dual spatial clustering algorithm based on field-theory.The primary contents of our paper can be concluded as follows:The characteristics of spatial relationship and potential scale between spatial objects are firstly analyzed based on the features of spatial data.In addition,a detailed comparison between spatial clustering and traditional clustering will be concluded,which can provide theoretical basis for the design and implementation of spatial clustering algorithms.Finally,the process of spatial clustering is also illustrated.According to the challenges of existing spatial clustering algorithms including uneven density and complicated spatial distribution,our paper introduces an innovative clustering method called Spatial Clustering with Multiple Density-Ordered Trees(SCMDOT)by using the hybrid clustering strategy.Motivated by the idea of the Density-Ordered Tree(DOT),SCMDOT has a combination of the idea of density peaks,graph theory,hierarchical clustering and density clustering.Furthermore,a series of MDOT can be successively generated from regions of sparse areas to the dense areas.In addition,both the intra-similarity of clusters and the inter-connectivity between clusters are taken into account to merge sub-clusters.The computational complexity of SCMDOT is approximately O(Nlog(N)).Experiments are utilized to demonstrate that our proposed method is effective and more reliable with regards to varied cluster sizes,shapes and densities.For the purpose of developing a spatial clustering algorithm considering both spatial proximity and attribute similarity,our paper makes improvements of the approach of handling attribute domain and proposes a modified dual spatial clustering algorithm based on field-theory(FTDSC).First of all,edge length constraints based on graph theory is employed for modeling the spatial proximity relationships among spatial objects.Then,inspired by the field theory in physics,the data field theory is utilized to describe the mechanism of spatial clustering in attribute domain.Moreover,aggregation force is used to express the degree of aggregation among the spatial objects from spatial data field and the iterative strategies are implemented to obtain different clusters according to the differences of attributes.The comparative experiments are made to further indicate that FTDSC algorithm can detect objects in the same cluster which are proximal in spatial domain and similar in attribute domain.By using the spatial clustering techniques mentioned above,the real-world application of catering industry agglomeration in Nanjing is studied and the detailed explanations of spatial clustering results are also given.The experiment shows that the effectiveness and practicability of the SCMDOT and FTDSC algorithm applying to real spatial dataset are validated,which is beneficial for better understanding of cluster features and the underlying causes in urban catering industry.