Research On Clustering Algorithms Of Spatial Objects And Moving Objects

Clustering analysis is one of the most important research fields in data mining, according to discovering descriptive objects and relational information in database, the goal of a clustering analysis is to partition the objects of a database into a set of close and independent clusters according to the criterion which is to maximize the similarity between the objects of each cluster and to minimize the similarity between the object of clusters. Nowadays, due to the data in the application database growing, clustering analysis have been widely applied in various of research fields and have become a very avtive research topic in the field of data mining. With the fast development of data collecting technology, computer network and database technology, data of complex data types increased dramatically, so explore novel data mining technologies are needed urgently to be applied to complex data types. This paper focus on the two problems research of clustering of spatial objects and trajectory clustering of moving objects form improving spatial objects clustering result, moving objects clustering result, improving clustering efficiency, and alleviating input parameters sensitivity. The main work are summarized as follows:Firstly, according to neighborhood, reverse neighborhood and local density of spatial which construct neighborhood-based local density, Afterwards, a novel clustering algorithm based on symmetric neighborhood of micro-clusters named BMSNC is proposed. BMSNC uses classical clustering algorithm to cluster original data sets to produce micro-clusters, then the center of each micro-cluster represent the micro-cluster, by symmetric neighborhood method to cluster the micro-clusters. The algorithm can effective and fast clustering analysis for large data sets, at the same time, It can make up the shortcoming that classical cluster can not distinguish small, dense, and adjacent clusters form large and sparse clusters, which makes the accuracy of clustering result improved greatly. Experimental results on real and synthetic data sets demonstrate that BMSNC is feasible and effective.Sencondly, the result of k-nearest neighbor clustering algorithm depends on the selection of distance metircs. The Euclidean distance which usually relates to all attributes. When feature weight parameters are intorduced to the distance formula, the result of clustering will depend on the weight values and accordingly can to improved by adjusting weight values. A clustering algorithm based on weight neighborhood named BWNC is proposed according to learning feature weights to improve the accuracy of clustering. BWNC add a feature parameter for each attribute, so different attribute can play a different role in the clustering. Mathematically it corresponds to a linear transformation for a set of points in the Euclidean space. It not only learned feature weights for each feature, but also weighted the contribution of each of the k neighbors according to their distance to the testing samples, that is, give greater weights to closer neighbors. So it can make the clustering results are more accurate and meaningful to the large extent through the test of experiment.Thirdly, a trajectory clustering algorithm based on symmetric neighborhood named BSNTC is proposed based on the existing trajectory clustering algorithm TRACLUS. BSNTC makes up the shortcome which TRACLUS can not distinguish small, dense, and adjacent trajectory clusters from large and sparse trajectory clusters due to using two global parameters Eps and MinLns. At the same time, BSNTC use a paramter k which can alleviate input parameter sensitivity. A series of experimental results show that the BSNTC improves the trajectory clustering result of TRACLUS while keeps the efficiency.