Research On The Fuzzy Multi-Level Co-Location Pattern Mining Algorithm Based On The Uniform Coefficient Of The Pattern Instance Distribution

With the rapid popularization of mobile Internet devices,the related research of spatial data mining has transitioned from the era of data scarcity to the era of data abundance.The massive spatio-temporal data provides an opportunity to obtain new knowledge and better understand complex geographical phenomena,and provides data support for solving urgent problems in the real world.The goal of mining spatial co-location patterns(SCP)is to discover subsets of spatial feature sets that have strong spatial correlations in the real world.This is an important technology for extracting and understanding tacit knowledge in spatial data,and it is also one of the important research directions in the field of spatial data mining.Due to the existence of both correlation and heterogeneity in spatial data,the spatial distribution of SCPs may appear globally in the entire study area,called a global co-location pattern(GCP),or only appear in a local area of the study area,called a region co-location pattern(RCP),based on this phenomenon,researchers propose multi-level co-location pattern mining(MLCPM).There are three main limitations of existing MLCPM methods.(1)When building the proximity relationship between instances,it is difficult for users to set an appropriate distance threshold to measure the proximity relationship between instances in a spatial dataset with heterogeneity.(2)When measuring the pattern frequency,the influence of the distance value between adjacent instance pairs on the pattern frequency is ignored,and the distribution characteristics of the pattern in space are not considered,resulting in unsatisfactory mining results.(3)The existing MLCPM methods take non frequent GCP as candidate RCP,the number of candidate RCP is too large,and regional co-location pattern mining(RCPM)relies on the instances of candidate RCP to construct the potential frequent regions of RCP,resulting in low overall mining efficiency.In order to solve the above three problems,firstly,this paper generates adjacent instance pairs of different feature types based on Delaunay triangulation method,and extracts two distance segmentation parameters to construct the fuzzy proximity relationship between instances.Then,the distance value of adjacent instance pairs is normalized by distance segmentation parameters,and a new pattern frequency measurement method is proposed based on fuzzy proximity relationship.Then,the distribution uniformity coefficient of pattern instances is defined to evaluate the distribution of patterns in space,and an MLCPM algorithm with pruning strategy is proposed.The algorithm considers both the frequency of patterns and the distribution of pattern instances in space,and can mine multi-level co location patterns accurately and efficiently.Finally,extensive experiments are carried out on real and synthetic data sets to comprehensively evaluate the mining efficiency,correctness and scalability of the proposed algorithm.