Clustering data stream using adaptive resonance theory

Data stream is a new data model on data mining in which data can be accessed orderly only once (or just a few passes) in an increasing number of data-intensive applications such as network monitoring, telecom call records on-line analysis, sensor networks, financial transactions, etc. In this model, data items arrive in continuous, rapid, time-varying, potentially infinite streams, which challenges data mining algorithm designing on efficiency, space consumption and the evolvement of the obtained model. This paper proposes ART (Adaptive Resonance Theory) neural networks, a class of architectures that self-organize stable clustering in real time in response to arbitrary sequences of input patterns, as a new tool in clustering data stream in view of its adaptivity between the new data and the achieved data model as well as its great efficiency of emulating algorithms. In the end experiments has been designed and the results present feasibility and validity of ART network in clustering data stream.