Spatial Co-Location Pattern Mining Based On The Fuzzy Neighborhood Relationship

In recent years,with the rapid development of spatial information technologies such as satellite positioning systems,geographic information systems(GIS)and remote sensing,a large amount of spatial data containing location information has been generated.Spatial data mining refers to the process of extracting implicit and user-interested spatial and non-spatial patterns,general features,rules and knowledge from spatial databases.A spatial co-location pattern is a subset of a set of spatial features that are related,and their instances coexist frequently in geographically adjacent spaces.Spatial co-location pattern mining is an important research direction of spatial data mining,which is widely used in many fields,including species distribution analysis,earth science,location-based services,public safety,urban facility planning and environmental management.Spatial co-location pattern mining is based on determining the spatial proximity relationships between instances to calculate the prevalence of a co-location pattern——participation index.The traditional spatial co-location pattern mining methods regarded the spatial neighborhood relationship as a deterministic concept,ignoring the proximity between instances,which makes the co-location pattern mining results lost accuracy and validity.In this paper,based on solving the above problems,a series of research work is carried out.The main work and contributions are summarized as follows:(1)To address the above problem,the fuzzy neighborhood relationship is used to describe the fuzziness of the spatial proximity relationship,and two important concepts are defined: fuzzy participation index and maximum proximity threshold.On this basis,two co-location pattern mining algorithms based on fuzzy proximity relationships are proposed: co-location pattern mining at a single proximity threshold and co-location pattern mining with the maximum proximity threshold.The former can excavate the co-location patterns at a given proximity threshold,and the accuracy and effectiveness are verified by comparing it with the traditional co-location pattern mining algorithms.The latter can discover all the co-location patterns and the maximum proximity threshold for each of them in the proximity threshold range.Given a proximity threshold,the prevalence of a co-location pattern can be directly judged based on the mining results of the latter.This paper also puts forward the improvement strategy of the latter.The experiments evaluate the performance of the algorithm and verify the efficiency of the improved strategy.(2)For the dynamic data sets,the algorithm of incremental mining of co-location patterns based on the fuzzy proximity relationships is proposed,which defines the prevalence measurement of the changed co-location pattern,makes full use of the mining results on the original data set and obtains the changed fuzzy proximity relationship to locally excavate the changed co-location patterns instead of re-excavating the co-location patterns on the entire updated dataset.The correctness and completeness of the algorithm are proved theoretically,and the efficiency of the algorithm is verified experimentally.(3)Because the traditional co-location pattern mining materialized the spatial proximity relationship according to a specific distance threshold,and takes time to repeatedly check the clique relationship and consumes memory to store the row instances,this paper explores a co-location pattern mining framework based on the combination of clustering and fuzzy proximity relationship,which uses clustering technology to materialize the spatial proximity between instances,without requiring the generation and storage of the row instances.It can greatly save the time and space complexity of generating co-location patterns.Three improved strategies of density peak clustering are proposed,and a new prevalence measurement of the co-location pattern is defined.And finally,an algorithm of co-location pattern mining based on density peak clustering and fuzzy proximity relationship is designed.The validity of the algorithm and the efficiency of generating co-location pattern are evaluated.