Mining Dynamic Spatial Co-location Pattern

Spatial co-location model mining as an important research direction in the field of data mining,Its purpose is to discover the coexistence relationship between spatial features and has a wide range of applications in many fields.However,there are two shortcomings in the current co-location pattern mining method.On the one hand,they ignore the existence of autocorrelation features that is not associated with surrounding features.For example,"cactus" and "Jerusalem artichoke" are two common plants in the desert,and it is easy to get prevalent pattern {cactus,Jerusalem artichoke} for the existing spatial co-location pattern mining frameworks,but biologists have determined that "Jerusalem artichoke" is a spatial autocorrelation feature so that above pattern is meaningless.On the other hand,although many spatial co-location pattern mining algorithms and their corresponding expansions have been proposed,many dynamic associations have not got enough attention.For example,the increasing amount of wolves leads to a decrease in the amount of sheep,and the decreasing amount of the wolves results in an increase of the sheep.In order to solve the above problems,in this paper,we propose a data preprocessing method based on k-means clustering to find and remove the spatial autocorrelation feature,and propose a spatial dynamic co-location model and its mining algorithm.Firstly,the research background,significance and current situation of spatial autocorrelation and spatial dynamic co-location mining are introduced.Secondly,a novel clustering algorithm is proposed in the data preprocessing stage.The algorithm combines the k-means algorithm with the clustering bias to find the spatial autocorrelation feature and remove it to avoid its influence on the mining result.The feasibility of the algorithm is verified by the simulation data.Thirdly,this paper proposes the definition and algorithm of spatial dynamic co-location mining,which obtains the neighbor set of newborn/death instances on the data set at multiple time points,and get the second-order candidate pattern and its table instance through this it,then through the join-based algorithm,we get the dynamic co-location frequent pattern of each order space.The algorithm can mine the meaningful pattern that is missing from the traditional way and make up the deficiency of the existing co-location pattern mining.Finally,the feasibility,validity and expansibility of the algorithm are verified on the simulation data and the real data.Finally,the paper summarizes and prospects the future work.