一种改进的局部离群数据检测算法

摘要

针对传统局部离群数据检测算法时间复杂度高、参数鲁棒性差的问题,在基于连接的异常因子(Connectivity based outlier factor,COF)算法的基础上,提出了一种基于聚类和密度的局部离群数据检测算法.利用聚类方法从原始数据集中筛选出候选离群数据集,来降低算法的时间复杂度;在进行数据对象之间距离计算时,引入信息熵的概念确定数据对象的离群属性,以提高算法的检测准确率.确定数据集的离群属性后,采用新的局部链接离群因子(Local connectivity based outlier factor,LCOF)度量候选离群数据集中数据的离群程度.此算法在保证检测准确率的前提下,降低了时间复杂度和检测准确率对参数的依懒性.仿真结果证明了所提方法的有效性和可行性.%To solve the problem of high time complexity and poor parameter robustness for the traditional local outlier detection algorithm,the local outlier improved detection algorithm considering clustering and density is proposed on the basis of improved COF.The algorithm uses clustering method to select candidate outlier datasets from the original data set so as to reduce the time complexity of the algorithm.Meanwhile,in order to improve the detection accuracy,the information entropy is introduced to determine the outlier attribute of the object when calculating the distance between data objects.After determining the outlier attribute of the datasets,using the new outlier factor LCOF to measure the outlier degree of the data in the candidate outlier datasets.The algorithm will reduce the time complexity and dependence of the detection accuracy on the parameter dependence.The simulation results verify the effectiveness and feasibility of the proposed method.