结合近邻思想和K-means的三支决策聚类方法  

Three-Way Decision Clustering Method Combining Nearest Neighbor Idea and K-means

作　　者：唐欣 TANG Xin(School of Mathematics and Information Science,North Minzu University,Yinchuan 750021)

机构地区：[1]北方民族大学数学与信息科学学院,银川750021

出　　处：《计算机与数字工程》2025年第2期314-319,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:62066001);宁夏自然科学基金项目(编号:2020AAC03217,2022AAC03238);北方民族大学研究生创新项目(编号:YCX21176)资助。

摘　　要：针对K-means算法随机选取聚类中心且易受极端值影响等问题,提出近邻思想和K-means的三支决策聚类方法。首先,利用样本点之间的关系得到密度最高的对象作为初始聚类中心,根据剩余样本点与初始聚类中心之间的近邻密度选取合适的聚类对象,同时更新聚类中心;接着从最远欧氏距离出发寻找n-1个聚类中心及其对应的聚类对象,得到二支K-means聚类结果。最后,结合三支决策和Q近邻思想,将上述结果进一步划分为核心域、边界域及琐碎域,得到三支K-means决策聚类结果。在UCI数据集和人工模拟数据集上分别进行试验,实验结果表明:相比于其他几种方法,该方法提高了聚类准确率,具有稳定性。Aiming at the problems of randomly selecting cluster centers and being easily affected by extreme values in K-means algorithms,a three branch decision clustering method based on the nearest neighbor idea and K-means is proposed.First⁃ly,using the relationship between sample points,the object with the highest density is obtained as the initial clustering center.Based on the density of neighbors between the remaining sample points and the initial clustering center,suitable clustering objects are selected,and the clustering center is updated at the same time.Then,starting from the farthest Euclidean distance,the article searches for n-1 clustering centers and their corresponding clustering objects,obtaining the results of two branch clustering.Final⁃ly,combining the three branch decision and nearest neighbor thinking,the above results are further divided into core domain,boundary domain,and trivial domain to obtain the clustering results of the three branch decision.Experiments are conducted on both the UCI dataset and the artificial simulation dataset,and the results show that compared to other methods,this method im⁃proves clustering accuracy and has stability.

关 键 词：K-MEANS算法 局部邻域密度 Q近邻 三支决策 三支聚类 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TP181[自动化与计算机技术—计算机科学与技术]