一种基于仿射传播的增强型流聚类算法

发布时间：2018-05-13 09:35

本文选题：流聚类 + 仿射传播　；参考：《西安交通大学学报》2017年03期

【摘要】：针对目前流聚类算法无法有效处理数据流离群点的检测和处理,以及增量式数据流聚类效率较低等问题,提出了一种基于密度度量的异常检测、删除的增强型仿射传播流聚类算法。在仿射传播流聚类算法的基础上,所提算法通过引进异常检测和删除机制改善了异常点对聚类精度、聚类效率的影响。利用仿射传播聚类实现在线数据流的聚类过程,同时检测数据漂移现象,即数据流分布特征随时间发生变化,并采用基于密度度量的局部异常因子检测技术(LOF)对储备池数据进行异常检测和删除处理,通过对当前类簇和处理过的储备池数据重聚类来重建动态数据流模型。在真实网络数据(KDD’99)上进行了实验,结果表明,所提算法不仅减少了重聚类构建动态模型的次数,改善了聚类效率,而且在同时考虑聚类精度、纯度和熵3种聚类评价标准下,均优于传统的仿射传播流聚类算法。
[Abstract]:Aiming at the problem that current flow clustering algorithm can not effectively deal with outlier detection and processing of data stream, and the efficiency of incremental data stream clustering is low, a density metric based anomaly detection method is proposed. Deletes an enhanced affine propagation flow clustering algorithm. Based on the affine propagation flow clustering algorithm, the proposed algorithm improves the effect of outlier points on clustering accuracy and clustering efficiency by introducing anomaly detection and deletion mechanisms. The affine propagation clustering is used to realize the online data flow clustering process, and the data drift phenomenon is detected at the same time, that is, the distribution characteristics of the data flow change with time. The local anomaly factor detection technique based on density metric is used to detect and delete the data of the reserve pool, and the dynamic data flow model is reconstructed by clustering the current cluster and the processed data of the storage pool. The experimental results on the real network data show that the proposed algorithm not only reduces the number of times of reclustering to construct dynamic model, but also improves the clustering efficiency, and considers the clustering accuracy at the same time. It is superior to the traditional affine propagation flow clustering algorithm under three clustering criteria of purity and entropy.
【作者单位】：西安交通大学软件学院;西安交通大学电子与信息工程学院;
【基金】：国家自然科学基金资助项目(61371087,61531013) 国家“863计划”资助项目(2015AA015702)
【分类号】：TP311.13

【相似文献】