Robust outlier detection using the instability factor
Robust outlier detection using the instability factor
复制标题
DOI：
10.1016/j.knosys.2014.03.001
复制
发表时间：
2014-06
期刊：
影响因子：
--
通讯作者：
J. Ha;Seulgi Seok;Jong-seok Lee
中科院分区：
文献类型：
--
作者：
J. Ha;Seulgi Seok;Jong-seok Lee
由于离群值检测适用于金融、电信、医疗和商业等各个领域，其重要性正在急剧增加。由于受到如此大的关注，导致了许多检测方法的发展，其中大多数属于基于距离的方法或基于密度的方法。然而，每种方法都有内在的弱点。前者很难检测到局部异常值，而后者存在低密度模式的问题。为了克服这些缺点，我们提出了一种新的检测方法，利用重心的概念引入数据点的不稳定因素。该方法通过控制其参数，可以灵活地用于局部和全局异常点检测。此外，它还提供了包含有关数据中簇的数量和大小的有用信息的不稳定性图。基于人工数据集和实际数据集的数值实验表明了该方法的有效性。
Since outlier detection is applicable to various fields such as the financial, telecommunications, medical, and commercial industries, its importance is radically increasing. Receiving such great attention has led to the development of many detection methods, most of which pertain to either the distance-based approach or the density-based approach. However, each approach has intrinsic weaknesses. The former hardly detects local outliers, while the latter has the low density patterns problem. To overcome these weaknesses, we proposed a new detection method that introduces the instability factor of a data point by utilizing the concept of the center of gravity. The proposed method can be flexibly used for both local and global detection of outliers by controlling its parameter. In addition, it offers the instability plot containing useful information about the number and size of clusters in data. Numerical experiments based on artificial and real datasets show the effectiveness of the proposed method.