基于加权自然近邻和测地距离的密度峰值聚类算法

打开文本图片集

关键词：聚类分析；密度峰值聚类；测地距离；加权自然近邻

中图分类号：TP301.6 文献标志码：A 文章编号：1001-3695（2026）01-011-0091-11

doi：10.19734/j.issn.1001-3695.2025.06.0200

Density peaks clustering algorithm based on weighted natural nearest neighbor and geodesic distance

Wan Fang，Wei Lili†，Liu Guojun （SchoolofMathematicsand Statistics，NingxiaUniversity，Yinchuan75oo21，China）

Abstract：DPC isaneffectiveand simpleclusteringmethodthatrequiresonlyoneparameterto identifyallclustercenters withoutiterativeprocesing.However，itstilexhibitslimitationswhenhandlingcomplexdatastructures.Firstly，theclusteringresultsaresensitivetothecutoffparameter.Secondly，theEuclideandistance-based metric tends toignorethegeometric structure ofmanifolddatasets，whichaffcts theaccuracyofthe clusteringresults.Toaddress these isues，this paperproposed adensitypeaks clustering algorithm basedon weighted natural nearest neighbor and geodesic distance（DPC-WNNN-GD）. This algorithmcouldanalysethelocalandglobal informationof thesamplescomprehensively，redefinethelocal densityby combing theweightednaturalnearestneighbors tobalancedensitydiferencesbetweensamples，andeliminatetheinfluenceof thecutoffparameter.Additionall，itreplacedEuclideandistancewithgeodesicdistancetobeteradapttothestructuralharacteristics ofmanifold datasets.ComparedDPC-WNNN-GDwithDPC algorithmandrelated improved algorithmsonsynthetic andreal datasets，experimental results demonstrate that DPC-WNNN-GD exhibits superior clustering performance.

Key Words：cluster analysis；density peaks clustering（DPC）；geodesic distance;weighted natural nearest neighbor

0 引言

流形数据[1]作为典型的密度分布不均匀数据，其内部隐含的复杂几何特征对数据挖掘提出了挑战。（剩余19579字）