郑州大学学报（工学版）2024，Vol.45Issue(5)：86-94,9.DOI:10.13705/j.issn.1671-6833.2024.02.002
基于边界剥离思想的全局中心聚类算法
Border-peeling Inspired Globally Central Clustering Algorithm
摘要
Abstract
The globally central clustering algorithms,such as k-means and spectral clustering,often suffer from the problem of local optima and difficulty in parameter setting with overlapping and adhesive clusters in the data distri-bution,which might greatly limits the effectiveness of globally central clustering algorithms in practical applications.To address this issue,a border-peeling inspired globally central clustering algorithm was proposed.Firstly,a one-step border peeling method was designed,which defines a locally distance-weighted density according to the reverse k-nearest neighbor relationships between sample points.The density value at the maximal point of the first-order difference of the density empirical distribution function was utilized as the threshold to divide the dataset into boundary and core sets.Then,the traditional globally central clustering algorithms were embedded to cluster the core set.Benefiting from the significant improvement in the overlapping of the core set,the embedding algo-rithms could converge to the true cluster centers easily.Finally,a boundary attraction algorithm was proposed,which could progressively amalgamate sample points from the boundary set,utilizing existing reverse k-nearest neighbor relationships,and commencing from the already categorized core set sample points.Compared with the currently iterative border peeling algorithms,the proposed algorithm had significant advantages in computational ef-ficiency.There was no additional complex termination conditions but only direct performs boundary partitioning u-sing a threshold.Furthermore,the global approach also exhibited stronger robustness local data densities were dif-ferent.In the experimental phase,three synthetic datasets and six real-world datasets were used to evaluate the al-gorithm's performance,parameter sensitivity,and time consumption,further validating the efficacy and practicality of this algorithm.关键词
全局中心聚类算法/边界剥离/簇重叠/反向k近邻/经验分布Key words
globally central clustering algorithm/border peeling/overlapping/reverse k-nearest neighbors/empiri-cal distribution分类
信息技术与安全科学引用本文复制引用
程明畅,敖兰,刘浏..基于边界剥离思想的全局中心聚类算法[J].郑州大学学报（工学版）,2024,45(5):86-94,9.基金项目
国家自然科学基金资助项目(12075162) （12075162）
数学地质四川省重点实验室开放基金资助(scsxdz2023-4) （scsxdz2023-4）
四川师范大学学科建设专项(XKZX2021-04) （XKZX2021-04）
