Scholarlink

<p>Clustering is a fundamental technique in unsupervised learning that identifies natural groupings within data. It has applications across data mining, machine learning, and pattern recognition. One of the major challenges in clustering is the absence of a universal criterion for selecting the most suitable method for different datasets. Therefore, various approaches have been proposed to enhance clustering accuracy and efficiency, among which ensemble and Hybrid Clustering Methods have gained significant attention. This paper introduces a novel Hybrid Clustering Method based on a modified K-Means algorithm and spectral graph partitioning. First, a set of base clusterings is generated using multiple executions of the K-Means algorithm. Then, a weighted graph representing the relationships between base clusters is constructed, and a normalized spectral clustering algorithm is applied to optimize the final clustering process. This approach reduces sensitivity to initialization and enhances clustering stability. Additionally, the proposed method improves the detection of both spherical and non-spherical clusters, thereby addressing a key limitation of traditional K-Means clustering. The proposed method was evaluated on 16 benchmark datasets using the Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), and F-measure as evaluation metrics, and compared with UCMean, Ensemble Clustering via Matrix Completion (ECMC), and Disciplined Convex–Concave Programming (DCCP). Experimental results demonstrate that the proposed method, on average, achieves improvements of approximately 8% in ARI, 6% in NMI, and 7% in F-measure compared with competing clustering methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A robust clustering approach: integrating spectral graph partitioning and adaptive K-Means

  • 期刊论文

  • 发表日期：

    2026年6月20日

查看全文
  • Sadegh Rezaei,
  • Razieh Malekhosseini,
  • S. Hadi Yaghoubyan,
  • Karamollah Bagherifard,
  • Samad Nejatian

摘要

Clustering is a fundamental technique in unsupervised learning that identifies natural groupings within data. It has applications across data mining, machine learning, and pattern recognition. One of the major challenges in clustering is the absence of a universal criterion for selecting the most suitable method for different datasets. Therefore, various approaches have been proposed to enhance clustering accuracy and efficiency, among which ensemble and Hybrid Clustering Methods have gained significant attention. This paper introduces a novel Hybrid Clustering Method based on a modified K-Means algorithm and spectral graph partitioning. First, a set of base clusterings is generated using multiple executions of the K-Means algorithm. Then, a weighted graph representing the relationships between base clusters is constructed, and a normalized spectral clustering algorithm is applied to optimize the final clustering process. This approach reduces sensitivity to initialization and enhances clustering stability. Additionally, the proposed method improves the detection of both spherical and non-spherical clusters, thereby addressing a key limitation of traditional K-Means clustering. The proposed method was evaluated on 16 benchmark datasets using the Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), and F-measure as evaluation metrics, and compared with UCMean, Ensemble Clustering via Matrix Completion (ECMC), and Disciplined Convex–Concave Programming (DCCP). Experimental results demonstrate that the proposed method, on average, achieves improvements of approximately 8% in ARI, 6% in NMI, and 7% in F-measure compared with competing clustering methods.