Scholarlink

<p>Density peaks clustering (DPC) algorithm has been a popular clustering algorithm in recent years. However, DPC and its variants usually require the manual selection of clustering centers on a decision map, a process that can be time-consuming. Furthermore, the accuracy of the clustering results is highly susceptible to the choice of the cutoff distance parameter. To address these shortcomings of DPC, we propose an automatic density peak clustering algorithm based on natural nearest neighbor and innovative sub-cluster merging(ADPC-NNSM). The local density is redefined according to the natural nearest neighbor relationship, eliminating the effect of the cutoff distance parameter. Then, an innovative approach is introduced to automatically select sub-cluster centers, avoiding the influence of human intervention. Subsequently, a distinctive merging strategy is utilized to combine these chosen sub-clusters until the end condition is satisfied. Theoretical analysis and experimental results obtained from twelve synthetic datasets and eight real-world datasets demonstrate that the overall clustering performance of our proposed ADPC-NNSM surpasses other compared algorithms.</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

Automatic density peak clustering algorithm based on natural nearest neighbors and innovative sub-cluster merging

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Yongqi Mi,
  • Xiangli Li,
  • Yalin Kou,
  • Zhibin Zhu

摘要

Density peaks clustering (DPC) algorithm has been a popular clustering algorithm in recent years. However, DPC and its variants usually require the manual selection of clustering centers on a decision map, a process that can be time-consuming. Furthermore, the accuracy of the clustering results is highly susceptible to the choice of the cutoff distance parameter. To address these shortcomings of DPC, we propose an automatic density peak clustering algorithm based on natural nearest neighbor and innovative sub-cluster merging(ADPC-NNSM). The local density is redefined according to the natural nearest neighbor relationship, eliminating the effect of the cutoff distance parameter. Then, an innovative approach is introduced to automatically select sub-cluster centers, avoiding the influence of human intervention. Subsequently, a distinctive merging strategy is utilized to combine these chosen sub-clusters until the end condition is satisfied. Theoretical analysis and experimental results obtained from twelve synthetic datasets and eight real-world datasets demonstrate that the overall clustering performance of our proposed ADPC-NNSM surpasses other compared algorithms.