Abstract: As an important kind of data source, data stream has received increasing attention.Data stream management systems and data mining based on data streams have also attracted much research interest.With dynamical gridpartitioning of the data space, distribution density of data streams is approximated, and based on which a density biased sampling method is presented.To test its efficiency, the proposed sampling method is applied to clustering data streams. Experimental results show promising applicability of the approach.

Key words: data streams, clustering, biased sampling