基于最小生成树与统计特征的层次聚类算法

打开文本图片集

中图分类号：TP181 文献标识码：A DOI：10.7535/hbkd.2026yx01006

Abstract：Toaddress the limitations of the Chameleon algorithm in terms of parameter sensitivity，noise robustness，and computationaleficiency，this studyproposed astatistical-MSTintegrated hierarchical clusteringalgorithm（SHCA）basedon the minimum spanning tree and statistical features.The minimum spanning tree was used to construct asparse graph， eliminatingmanualparameterintervention，andtheglobaloptimalityoftheminimumspanning tree wasused toavoid false crossclusterconnections.Thedynamic statistical mergingstrategy was designed to filterthe noisecombined withthe local distance threshold，andthesubclusters were mergediterativelythrough theinterclusterconnectivitytesttoensure theintra cluster compactnessand inter cluster separation.Experiment on 20synthetic datasetsand1real-world datasetswas conducted.The result shows thatthe proposed SHCA algorithm outperforms existing methods in clustering performance；In cases whereperformance degradationisobservedoncertaindatasets，theanalysis revealsthat manifoldoverlapistheprimary contributing factor.Overall，SHCA significantlyenhances clustering accuracyandresult stability，providing somereferencefor subsequent research on clustering of large-scale and complex manifold data.

Keywords：artificial intellgence theory；clustering；hierarchicalclusteringalgorithm；minimum spanning tree；dynamicstatistical merging strategy

一般而言，聚类是指将无标签数据集划分为若干类别，使得类内数据相似度高、类间数据相似度低的过程，是一种无监督的机器学习分类方法[1-2]。（剩余13343字）