基于自适应的高对比性子空间的高维离群点检测

Mining outlier in high dimensional space based on adaptive high contrast subspace

基于子空间解决高维离群点挖掘的问题已经引起人们的广泛关注,现有方法存在的主要问题是难以选取合适的子空间且选取计算量大、阈值等参数设置困难等。这些影响了检测精度和检测效率。利用高对比度子空间选取方法解决子空间选取问题,利用自适应方法解决阈值参数的确定问题,据此提出自适应的高对比性子空间离群点检测方法(AHiCS)。该方法利用统计检验算法选取高对比性子空间,在高对比性的子空间里自适应计算离群点得分,提高了离群点检测的精度与效率。理论和实验表明,该方法可以有效地挖掘高维离群点。

Detecting outlier in high dimensional space based on subspace has aroused extensive attention, the main problems that existing methods have are：difficult to select the appropriate subspace and set the threshold parameter,calculation and se- lection cost much time. These affect the detection accuracy and efficiency. This paper used high contrast subspace to solve the problem for selecting subspace, and used adaptive method to solve the threshold parameter, so it proposed the outlier detection method based on adaptive high contrast subspace （AHiCS）. The method used statistical test algorithm of selecting the high con- trast subspace and calculate outlier score under the finding subspace. It improved the accuracy and efficiency of the outlier de- tection. Theoretical and experimental results show that the method can effectively detect outliers.