摘要
为识别混合属性数据集中的离群点,提出了一种基于共享最近邻的离群检测算法,通过计算增量聚类结果簇间的共享最近邻相似度,不但能够发现任意形状的簇,还可以检测到变密度数据集中的全局离群点。算法时间复杂度关于数据集的大小和属性个数呈近似线性。在人工数据集和真实数据集上的实验结果显示,提出的算法能有效检测到数据集中的离群点。
This paper introduced an outlier detection algorithm based on the shared nearest neighbor clustering in order to detect the outliers with the mixed attributes.The algorithm calculated the shared nearest neighbor similarity measure between result clusters caused by the incremental clustering.It could not only find the arbitrary shape clusters but also identify the global outlier in large and high-dimensional dataset with different density.Presented approach had nearly linear time complexity with the number of attributes and the size of dataset which results in good scalability.
出处
《计算机应用研究》
CSCD
北大核心
2012年第7期2426-2428,2453,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61163017)
郑州轻工业学院博士科研基金资助项目(2010BSJJ039)
河南省科技攻关资助项目(122102210125)
河南教育厅自然科学基础研究计划资助项目(12B520051)
关键词
共享最近邻
离群检测
任意形状簇
混合属性
shared nearest neighbor（SNN）
outlier detection
arbitrary shape cluster
mixed attributes
