基于模糊邻域熵的离群点检测方法

Fuzzy neighborhood entropy⁃based outlier detection

离群点检测(又称异常点检测)是数据挖掘领域中一个重要的研究方向,其目的是找出显著区别于其他数据的数据点.针对基于传统粗糙集理论的离群点检测方法存在忽略样本的模糊性和邻域关系等问题,利用模糊邻域粗糙集弥补经典粗糙集的不足,并结合熵的不确定性,提出一种新的基于模糊邻域熵的离群点检测方法.首先,采用模糊邻域半径和混合模糊相似度构造模糊邻域近似空间;然后,定义一种特定的模糊邻域组合熵和相对模糊邻域组合熵来构建模糊邻域离群度,进而定义基于模糊邻域熵的离群因子实现离群点检测,并设计了基于模糊邻域熵的离群点检测算法(FNEOD).最后,将FNEOD算法与主要的离群点检测算法进行比较.实验结果表明,该方法具有较好的有效性和适应性.

Outlier detection(also known as anomaly detection)is an important research direction in the field of data mining,with the aim of identifying data points that are significantly different.In response to the problem of neglecting the fuzziness and neighborhood relationships of samples in outlier detection methods based on traditional rough set theory,this paper uses fuzzy neighborhood rough sets to compensate for the shortcomings of classical rough sets,and combines the uncertainty of entropy to propose a novel outlier detection method based on fuzzy neighborhood entropy.Firstly,a fuzzy neighborhood approximation space is constructed using fuzzy neighborhood radius and mixed fuzzy similarity.Then,a specific fuzzy neighborhood combination entropy and a relative fuzzy neighborhood combination entropy are defined to construct fuzzy neighborhood outliers.Furthermore,an outlier detection algorithm based on fuzzy neighborhood entropy(FNEOD)was designed by combining the outlier factor based on fuzzy neighborhood combination entropy.Finally,the FNEOD algorithm is compared with the main outlier detection algorithms.The experimental results show that this method has good effectiveness and adaptability.