衰减窗口中的不确定数据流聚类算法

针对现实不确定数据流具备分布非凸性和包含大量噪声等特点,提出不确定数据流聚类算法Clu_Ustream(clustering on uncertain stream)来解决对近期数据进行实时高效聚类演化问题.首先,在线部分利用子窗口采样机制采集滑动窗口中的不确定流数据,采用双层概要统计结构链表存储概率密度网格的统计信息;然后,离线聚类过程中通过衰减窗口机制弱化老旧数据的影响,并定期对窗口中的过期子窗口进行清理;同时采用动态异常网格删除机制有效过滤离群点,从而降低算法的时空复杂度.在模拟数据集和网络入侵真实数据集上的仿真结果表明,Clu_Ustream算法与其他同类算法相比具有较高的聚类质量和效率.

Clustering algorithm for uncertain data stream based on damped sliding window

In view of the fact that uncertain data stream had the characteristics of non-convex distribution and contained a lot of noise, this paper proposed an algorithm Clu_Ustream for clustering uncertain data stream which solved the problem of real-time and efficient clustering evolution for recent data. Firstly, in the online part, Clu_Ustream used the sub window sampling mechanism to collect the uncertain stream data in the sliding window. Moreover, it used a double-layer summary statistical structure linked list to store the statistical information of the probability density grids to improve the processing efficiency. Secondly, in the off-line part, it used the damped window mechanism to weaken the influence of old data and deleted regularly the expired sub windows to ensure the effectiveness of clustering. In addition, it developed a dynamic abnormal grids deletion mechanism to filter most of outliers in order to dramatically improve the space and time efficiency. The experimental results on the synthetic and real datasets show that Clu_Ustream has superior clustering quality and efficiency than other similar algorithms.