Research On Varying-Density Spatial Clustering In High-Dimensional Data

With the rapid development and wide application of information technology, the high-dimensional data are stored redundantly. How to discover knowledge from the desert of data has become the most pressing requirement.As an important and active branch of data mining, clustering attemp to discover valuable knowledge from huge unknown database. Furthermore, clustering has been broadly applying in pattern recognition, image processing, market research, life science and so on. However, the real data are often high-dimensional, sparse and noisy; especially for its distribution of varying density, which makes the traditional clustering algorithms fall in predicament. Therefore, the design of efficient model and algorithm for clustering high-dimensional and varying density data becomes the core of this study.The main works are as follows:(1) The ideas, principles, implements, advantages and disadvantages of kinds of clustering methods are firstly probed into and analyzed deeply.(2) Aiming at the advantages and deficiencies of density-based clustering algorithm, a Hierarchical Tree (H-Tree) model is introduced to describe sub-cluster information, then the idear of density-based clustering is adopted to detect clusters and form DCHT algorithm, which not only possesses the advantages of density-based clustering, but also removes its defects resulted in the rough and inadequate design of structure.(3) In the face of the actuality that traditional clustering methods are difficult to deal with varying density spatial clustering, an improved H-Tree model is introduced to pre-treat the database and get the distribution information; then the local parameter tune dynamically and self-adaptively. Afterwards, a new algorithm SVC utilizes this local parameter to solve the problems of varying density spaticial clustering well.Both theoretical analysis and experimental results confirm and demonstrate the efficiency and effectiveness of the above two algorithms.