基于投票理论的三支聚类分析

Three-Way Clustering Analysis Based on Voting Theory

摘要: 目前，大多数聚类方法是二支聚类，即每个对象要么属于一个类，要么不属于一个类，聚类结果具有清晰的边界。然而，将某些不确定的对象强制分配到某个类簇中会降低聚类结果精度。而三支聚类是一种重叠聚类方法，它采用核心域和边界域来表示每个类簇，较好地处理了含有不确定性信息对象的类别归属问题，同时有效地降低了决策风险。本文主要介绍一种三支决策聚类模型，并给出基于k-means的三支决策聚类算法作为实例进行分析。首先，通过已有的聚类算法得到相同数据集的不同二支决策聚类结果，然后对聚类成员的类簇标签进行匹配，最后制定投票规则确定样本的类簇归属。通过实验结果分析，本文所提出的聚类方法的聚类结果在各聚类性能度量指标上有了明显的提升。

Abstract: At present, most of existed clustering methods are two-way clustering which are based on the as-sumption that a cluster must be represented by a set with crisp boundary. However, assigning uncertain objects into a certain cluster will reduce the accuracy of clustering results. Three-way clustering is an overlapping clustering which describes each cluster by core region and fringe region. It handles the category problem of uncertain objects and reduces the decision risk effectively. This paper mainly introduces a model of three-way clustering, and gives three-way clustering algorithm based on k-means as an example for analysis. Firstly, different clustering results of the same data set are obtained by ensemble clustering. Then, the label matching method is used. Finally, the cluster of objects is determined according to the voting rules. Through the analysis of experimental results, it is verified that the effect of the clustering method has been significantly improved.