统计与决策2025，Vol.41Issue(11)：49-54,6.DOI:10.13546/j.cnki.tjyjc.2025.11.008
基于核函数的自适应谱聚类与聚类个数确定方法
Adaptive Spectral Clustering and Determination of the Number of Clusters Based on Kernel Function
摘要
Abstract
This paper first compares the effects of different kernel functions on spectral clustering(SC),then constructs three kinds of adaptive spectral clustering(ASC)based on the k-nearest neighbor idea,also constructing the differential determination of the number k* of clusters based on the eigenvalues of Laplace matrix,and finally constructs the clustering-KNN algorithm and uses its instability to determine k*.Numerical simulation results are shown as the following:The kernel function SC has a wide range of adaptations;it is effective for non-convex data sets under a suitable kernel function,and Gaussian kernels are recom-mended.The Gaussian kernel is significantly affected by the global parameters,and the three types of ASC are robust to the neigh-bor parameters.Determining k* based on the instability of the clustering-KNN algorithm has a wider range of application than the statistical method;for non-convex datasets,it is recommended to select kernel function spectral clustering as the basic clustering algorithm.The random sampling method is robust to the sampling number m.When the proportion of mis relatively high,it is ap-proximately similar to the self-sampling method.关键词
核函数/谱聚类/聚类个数/不稳定性Key words
kernel function/spectral clustering/number of clusters/instability分类
数学引用本文复制引用
王丙参,魏艳华,李旭..基于核函数的自适应谱聚类与聚类个数确定方法[J].统计与决策,2025,41(11):49-54,6.基金项目
甘肃省高校教师创新基金项目(2025B-156) （2025B-156）
山西省自然科学基金青年项目(202203021222223) （202203021222223）
