基于粒球的三支聚类方法

Three-way clustering based on granular-ball

针对硬聚类中样本均被强制划分到一个类簇,没有考虑信息具有不确定性而导致决策风险提高的问题,将粒球与三支聚类结合,提出了一种基于粒球的三支聚类算法.通过对每一个数据簇建立一个粒球模型,以自适应的方式为每个粒球寻找其近邻粒球.根据近邻粒球的情况来将粒球划分为稳定域与活动域,以赋予不同的权重将稳定域与活动域中的样本数据进行更新迭代.稳定域被认定为该类的核心域,活动域为该类的边间域.采用不同的UCI数据集进行实验,实验结果表明:所提算法较好的解决了不确定性信息对于决策风险提高的问题,验证了该算法的有效性.

In order to solve the problem of traditional clustering where samples are forcibly divided into one cluster without considering information uncertainty,thereby increasing decision risk.In this paper,a three-way clustering algorithm based on granular-ball is proposed by combining granular-ball with three-way clustering.The algorithm establishes a granular-ball model for each data cluster,and finds its neighboring granular-ball for each granular-ball in an adaptive way.The granular-ball is divided into stable area and active area according to the situation of the neighboring granular-ball,and then the sample data in the stable area and active area are updated and iterated by giving different weights.The stable area is identified as the core region of this class,and the active area is the fringe region of this class.Experiments are conducted using different UCI datasets,and the experimental results show that the proposed algorithm effectively solves the problem of increasing decision risk due to uncertain information,verifying the effectiveness of the algorithm.