earticle

영어

In recent years, clustering data streams has been actively proposed in the field of data mining. In real-life domains, clustering methods for data streams should effectively monitor the continuous change of a data stream with respect to all the dimensions of the data stream. In this paper, a clustering method with frequency prediction of data elements is proposed. The incoming statistics of data elements in the monitoring range are maintained. For the range of elements with high density, the range is partitioned to detect the detailed boundary of clusters. To identifying the recent change of a data stream quickly, the support of elements is carefully monitored and predicted to determine partitioned ranges to become clusters. Considering the change of the data stream, a threshold is adaptively controlled by a prediction mechanism. By predicting the change of supports, the on-going change of a data stream can be reflected in real-time. The proposed method is comparatively analyzed by a series of experiments to identify its various characteristics.