Abstract Three-way clustering provides an effective framework for clustering of data in the presence of uncertain, imprecise and incomplete data. In this article, we used ideas inspired from two commonly used spatial filters from image processing called minimum and maximum filters to construct a three-way clustering approach named RE3WC and explore its application in outlier detection. A three-way cluster is based on a pair of sets known as the core and support of a cluster. Given the results of a hard clustering algorithm in the form of hard clusters, RE3WC uses reduction and elevation operations to shrink and enlarge a hard cluster into the core and support of a three-way cluster. The two sets are then used to obtain the three regions of a three-way cluster namely, inside, outside and partial regions. Experimental results on CHAMELEON and other similar datasets indicate that the RE3WC can detect an additional 2.5% to 4.6% of objects as outliers that went undetected with clustering algorithms that detect outliers. The RE3WC results in more compact and precise clusters when applied on top of clustering algorithms that only provide partitioning of the data. Finally, RE3WC produces comparable results to some of the notable approaches such as LOF, LoOP, ABOD and IF.

中文翻译：

---

一种空间滤波启发的三路聚类方法，并应用于异常值检测

摘要 三向聚类为存在不确定、不精确和不完整数据的情况下的数据聚类提供了有效的框架。在本文中，我们使用灵感来自图像处理中两个常用的空间过滤器（称为最小过滤器和最大过滤器）的想法来构建名为 RE3WC 的三向聚类方法，并探索其在异常值检测中的应用。三路集群基于一对称为集群的核心和支持的集合。给定硬聚类形式的硬聚类算法的结果，RE3WC 使用归约和提升操作将硬聚类缩小和放大为三路聚类的核心和支持。然后使用这两组来获得三路集群的三个区域，即内部、外部和部分区域。CHAMELEON 和其他类似数据集的实验结果表明，RE3WC 可以检测到额外 2.5% 到 4.6% 的对象作为异常值，而这些异常值是用检测异常值的聚类算法检测不到的。当应用于仅提供数据分区的聚类算法之上时，RE3WC 会产生更紧凑和更精确的聚类。最后，RE3WC 产生的结果与一些著名的方法（如 LOF、LoOP、ABOD 和 IF）相当。