摘要
聚类是一种典型且重要的数据挖掘方法,但现有聚类算法大多需要人为指定聚类的数量,并且聚类结果对参数敏感.针对上述不足,本文提出一种基于子博弈完美均衡的启发式聚类算法(Heuristic Clustering algorithm based on Sub-game Perfect Equilibrium,HCSPE).该算法充分挖掘数据点自身的分布特征信息,通过启发式方法得到自适应的参数值,从而使数据点局部密度属性值的得出具有客观性和普适性,降低了聚类结果对参数的敏感性.基于博弈的思想,综合局部密度和相对距离两个属性形成数据点的竞争力,依靠竞争机制完成聚类数量的自动计算以及聚类中心的确定.在多个规模和类型均不相同的数据集上的实验结果表明,本文所提出算法的性能指标整体优于其他算法,并且聚类结果更符合客观所需.
Clustering is a typical and important data mining method,but most of the existing clustering algorithms need to specify the number of clusters artificially,and the clustering results are sensitive to parameters.To address the above shortcomings,this paper proposes a heuristic clustering algorithm based on sub-game perfect equilibrium(HCSPE).The algorithm fully exploits the information of the distribution characteristics of data points themselves and obtains the adaptive parameter values by heuristic methods,so that the local density attribute values of data points are derived with ob⁃jectivity and universality,and the sensitivity of clustering results to parameters is reduced.Based on the idea of game,the two attributes of local density and relative distance are integrated to form the competitiveness of data points,and the auto⁃matic calculation of the number of clusters and the determination of cluster centers are completed by relying on the competi⁃tion mechanism.The experimental results on several data sets of different sizes and types show that the performance index⁃es of the proposed algorithm are better than other algorithms in general,and the clustering results are more in line with the objective requirements.
作者
常璐瑶
牛新征
罗涛
钱早国
CHANG Lu-yao;NIU Xin-zheng;LUO Tao;QIAN Zao-guo(School of Computer Science and Engineering,University of Electronics Science and Technology of China,Chengdu,Sichuan 611731,China)
出处
《电子学报》
EI
CAS
CSCD
北大核心
2024年第3期740-750,共11页
Acta Electronica Sinica
基金
国家自然科学基金(No.62272087)
四川省科技计划项目(No.2021YFS0391)。
关键词
博弈论
竞价机制
子博弈均衡
启发式算法
聚类
game theory
bidding mechanism
sub-game equilibrium
heuristic algorithm
clustering
