牛顿-拉夫逊优化驱动的自适应密度峰值聚类

打开文本图片集

中图分类号：TP181 文献标志码：A 文章编号：1001-3695（2026）04-020-1139-09

doi：10.19734/j.issn.1001-3695.2025.08.0281

Newton-Raphson optimization-driven adaptive density peak clustering

Wei Xiuxila，Li Kang1b，Huang Huajuanlat， Zhou Yongquan 1a，1c. ，Pang Qiuben²（1.a.Schoialei，lfsdeooiKbofICDesignAlisiUstetfatesfsfiMedical University，Nanning ，China）

Abstract：Toaddress thechalenges of densitypeak clustering （DPC），including manual setting ofcutof distance parameter，sensitivitytonoise，andlimitedadaptabilityto non-uniformdatadistributions，thisstudydesigneda Newton-Raphson optimization-drivenadaptive densitypeaks clustering method（NRO-ADPC）.This method establishedanadaptive parameter optimizationmechanismbasedonNewton-Raphsonoptimizationalgorithm toautomaticallydeterminecutoffdistancethrough gradient-guidedsecond-orderoptimization，eliminating manual parameter tuning dependency.This studyconstructedamultiobjectiveoptimizationfunctionintegratingdensitycontinuityclusterseparation，andpeakdistinctness，combinedwithadaptive weighting mechanismtoenhancealgorithmrobustnessagainstnoise.Thestudyadopteda hybriddensityestimationstrategy combiningfixedandadaptivebandwidthtoadapttolocaldistributionvariationswhilemaintainingglobalconsistencyExperiments on5 synthetic and5 real-world datasets demonstrate that NRO-ADPC achieves over 99% average accuracy on synthetic datasets and improves accuracy by over 20% compared to traditional DPC on real-world datasets. Wilcoxon signed-rank test shows that NRO-ADPC achieves statistically significant advantages over comparison algorithms including MDPC + in ACC and NMI metrics （ P<0.05 ），exhibiting superior performance when processing complex data with high noise，multi-scale features，and non-uniform distributions.

Key words：densitypeak clustering；Newton-Raphson optimization；adaptive parameteroptimization；noiserobustness；nonuniform distribution

0 引言

聚类在数据分析中发挥着至关重要的作用，它能够自动将未标记数据组织成有意义的组。（剩余19865字）