An algorithm for clustering uncertain data streams over sliding windows

The existing algorithms for clustering data streams with uncertainty can not analyze recent data in detail. In this paper, we propose SWCUStreams (Clustering Uncertain Data Streams over Sliding Windows) to cluster uncertain data streams, which can obtain the distribution character of recent data by maintaining the Exponential Histogram of Uncertainty Cluster Feature (EHUCF). SWCUStreams adopts the clustering framework of CluStream. In the online micro-cluster phase, Uncertainty Temporal Cluster Feature (UTCF) is defined to describe the uncertainty tuples. Based on the Uncertainty Temporal Cluster Feature (UTCF), Exponential Histogram of Uncertainty Cluster Feature is proposed to store the distribution character of recent data as well as used to dynamically delete expired records included in EHUCF by associating with UTCF. In the offline macro-cluster phase, the final clustering results will be generated according to the statistic information of Exponential Histogram of Uncertainty Cluster Feature (EHUCF) by UK-means algorithm. The experimental results over different types of data sets show that the cluster quality of SWCUStreams is higher.