Research On Data Stream Faced Clustering Algorithm

Clustering analysis is an important task in data mining field. Data stream clustering has became a challenging and meaningful topic over the topic past few years. Comparing with the traditional static data, these data are successive, dynamic, variational and boundless. These characteristic make data stream clustering algorithm become difficult. The problem will become even more challenging when the data is high-dimensional and non-linear.In this paper, an effective data clustering algorithm over high-dimensional and non-linear data stream is presented, this algorithm also adapts to the stream's evolutionary changes. Using the kernel method that is capable to handle non-linear problem, an innovative 2-tier stream clustering structure is proposed. First-tier captures the similar data in the stream by partitioning it into segments, using a kernel-based novelty detection method. Second-tier projects the high-dimensional and non-linear data to be low-dimensional.The correlative experiment demonstrates that the algorithm has good applicability, effectiveness and scalability. The algorithm is suitable for dealing with high-dimensional and non-linear data stream.