The Clustering And Detection Of Clusters’ Boundary In Real-time Data Streams

Mining real-time data stream is a very importance research hotspot in the area of datamining. In the modern society with the rapid development of computer technology and thepopularization of the computer application, real-time data stream arises at the historic moment. Itis characteristic of timeordered, rapid change, concept drift, tremendous data, potential infiniteand so on. Mining real-time data stream exactly mining the unknown and interested knowledgefrom these huge data. As an important research field of mining real-time data stream, thetechnique of clustering analysis and its boundary detection get more attention from researchers,and it has been a very active research topic in mining real-time data stream.In this thesis, grid-based clustering method based on damped window is adopted to discoverthe clusters and their boundary from real-time data stream called GDBOUND. GDBOUNDcalculate each grid’s density and the similarity between grids, finds the boundary from theclustering results, improves the performance of the algorithm. GDBOUND has the ability toreal-time response clustering request, comparing the clustering and boundary results at differenttime, it can implements the data stream evolving analysis.The algorithm has been implemented with VC6.0platform. The experimental results showthat the algorithm GDBOUND effectively discovers clusters and their boundaries of arbitraryshapes, any size and different density in data streams. GDBOUND has the advantages ofclustering and boundary, low memory consumption and performance.