一种基于偏离的局部离群点检测算法  

Deviation-based local outlier detection algorithm

作　　者：周世波[1,2] 徐维祥[1] 

机构地区：[1]北京交通大学,北京100044 [2]集美大学航海学院,厦门361021

出　　处：《仪器仪表学报》2014年第10期2293-2298,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61272029);李尚大集美大学学科建设基金(ZC2011018)资助项目

摘　　要：针对现有的局部离群点检测算法对数据对象不加分区,致使计算复杂度高的问题,提出了一种基于偏离的局部离群点检测算法。该算法首先对数据集进行分区,将可能存在的局部离群点与其紧邻的簇划分到一个数据块中,然后在每个数据块内,根据离散系数刻画各个数据对象的偏离度,从而求得每个数据对象在其所属的数据块内的局部偏离因子,发现可能存在的局部离群点。理论分析和实验结果表明,该算法具有良好的识别局部离群点的能力,检测的准确率和时间效率均优于经典的LOF算法。Aiming at the problem that existing local outlier detection algorithm does not perform partition of data objects, which results in high computational complexity, a deviation-based local outlier detection algorithm is introduced. The algorithm first divides the data set into sections, puts the potential outliers and their near clusters into a local neighbourhood, then in each local neighbourhood the local de- viation factor of each data object is described with the variation coefficient, as a result, the local variation of each data object in its be- longed data block is obtained, and the potential local outliers can more likely be found. The theoretic analysis and experiment results in- dicate that the proposed method has good local outlier recognition ability, the accuracy and time efficiency of local outlier detection are better than those of classical local outlier foitor（LOF） algorithm.

关 键 词：聚类 局部离群点检测 局部偏离因子 离散系数 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置] TH701[自动化与计算机技术—控制科学与工程]