计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (3): 94-102.DOI: 10.3778/j.issn.1002-8331.2007-0205

• 理论与研发 • 上一篇    下一篇

结合鲸鱼优化算法的自适应密度峰值聚类算法

王芙银,张德生,张晓   

  1. 西安理工大学 理学院,西安 710054
  • 出版日期:2021-02-01 发布日期:2021-01-29

Adaptive Density Peaks Clustering Algorithm Combining with Whale Optimization Algorithm

WANG Fuyin, ZHANG Desheng, ZHANG Xiao   

  1. College of Science, Xi’an University of Technology, Xi’an 710054, China
  • Online:2021-02-01 Published:2021-01-29

摘要:

针对密度峰值聚类算法(DPC)的聚类结果对截断距离[dc]的取值较为敏感、手动选取聚类中心存在着一定主观性的问题,提出了一种结合鲸鱼优化算法的自适应密度峰值聚类算法(WOA-DPC)。利用加权的局部密度和相对距离乘积的斜率变化趋势实现聚类中心的自动选择,避免了手动选取导致的聚类中心少选或多选的情况;考虑到合理的截断距离[dc]是提高DPC算法聚类效果的重要因素,建立以ACC指标为目标函数的优化问题,利用鲸鱼优化算法(WOA)有效地寻优能力对目标函数进行优化,寻找最佳的截断距离[dc];利用人工合成数据集与UCI上的真实数据集对WOA-DPC算法进行测试。实验结果表明,该算法在FMI、ARI和AMI指标上均优于DPC算法、DBSCAN算法以及K-Means算法,具有更好的聚类表现。

关键词: 密度峰值聚类算法, 鲸鱼优化算法, 聚类中心自适应, 截断距离

Abstract:

To solve these problems of the Density Peaks Clustering algorithm(DPC) that the clustering results are more sensitive to the cutoff distance [dc], as well as the clustering centers selected manually are subjective, an adaptive Density Peaks Clustering algorithm combining with Whale Optimization Algorithm(WOA-DPC) is proposed. The selection of clustering center is automatically realized according to the slope variation trend of the weighted product of the local density and the relative distance, which avoids the situation that the number of the clustering centers selected by manual operation is larger or smaller. The reasonable cutoff distance [dc] is an important factor to improve the clustering result of DPC. An optimization problem with the objective function being the ACC index is established. The objective function is optimized by using the effective optimization ability of the Whale Optimization Algorithm(WOA) to find the best cutoff distance [dc]. The proposed WOA-DPC is tested with the artifical datasets and the real datasets on UCI. Experimental results show that the proposed algorithm outperforms DPC, DBSCAN and K-Means in terms of FMI, ARI and AMI indicators with better clustering performance.

Key words: density peak clustering algorithm, whale optimization algorithm, cluster center adaptive, cutoff distance