基于水波传播特性的半监督密度聚类算法

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）08-011-2329-06

doi：10.19734/j. issn.1001-3695.2025.01.0016

Semi-supervised density clustering algorithm based on water wave propagation characteristics

Feng Yanchaola，Wang Jiela，LiYatinglb，Cai Jianghuila.2，Yang Haifenglat （1..ScolfoerSece&TohlofectinEgng，nUiesifSee & TaiyuanO30024，China；2.ScholofComputercience&Technology，North UniversityofChina，TaiyuanO3O51China）

Abstract：Semi-superviseddensityclusteringoptimizes theclustering performance by integrating supervised informationand densityfeatures.However，whenthedataisunevenlydistributed，thetraditionalsemi-superviseddensityclusteringalgorithm isoftendificulttocapturelocaldensitydiferencesduetotheglobaldensitythresholdsettingandinsufcientuseofcostraint information，resulting inlimitedrecognitionabilityofsparseregionalclusters.Toddressthischallnge，thispaperproposed a semi-upervised densityclustering algorithm based on water wave propagation characteristics（SDR）.Inspiredbythe“constantwavecenter”and“amplitudeattenuation”characteristicsof water waves，SDR simulatedthe tendencyof clustercenter stabilityanddensitydecreasingfromcentertoboundary，andidentifiesinitialsubclustersbylocaldensitydiferences.Thealgorithmfurther utilized the neighborinfluencetoassigntheremainingpoints tothesubclusters，and dynamicallmerged these subelustersbasedonthelabelatachmentdegreeTheexperimentalresultson5syntheticdatasetsand7realdatasetsconfim the validity of SDR algorithm in sparse region clustering task.

KeyWords：semi-supervised learning;density-based clustering；water wave propagation；constrained optimizatior

0 引言

聚类是一种数据分析技术，它将数据集中的样本划分为若干个簇，确保同一簇内的样本具有较高的相似性，而不同簇之间的样本相似性较低，从而揭示数据集的内在结构和模式。（剩余15576字）