Scholarlink

Clustering algorithms are essential tools in data mining, applied across various domains. Among them, DBSCAN has gained significant popularity due to its ability to identify clusters of arbitrary shapes without requiring the number of clusters to be specified in advance. However, as fairness in clustering becomes an increasingly critical concern, traditional DBSCAN algorithms face a lack of fairness consideration, which may result in biased clustering outcomes. To solve this issue, this paper proposes an improved clustering algorithm—FairDBSCAN (FDBSCAN). Based on the traditional DBSCAN framework, FDBSCAN introduces the novel concept of Fair Directly Density-Reachable and incorporates an innovative dynamic fairness-aware mechanism. The algorithm leverages a dynamic adjustment strategy to balance clustering performance and fairness during the cluster expansion process. Additionally, FDBSCAN employs a probabilistic reassignment strategy to ensure that unassigned points are represented fairly. Experiments conducted on multiple datasets demonstrate that FDBSCAN significantly improves fairness while achieving clustering results comparable to those of traditional methods.

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

FDBSCAN: A Dynamic Fairness Optimization Approach for Density-Based Clustering

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Junjie Zheng,
  • Zhijing Yang,
  • Yuxin Li,
  • Hui Zhang

摘要

Clustering algorithms are essential tools in data mining, applied across various domains. Among them, DBSCAN has gained significant popularity due to its ability to identify clusters of arbitrary shapes without requiring the number of clusters to be specified in advance. However, as fairness in clustering becomes an increasingly critical concern, traditional DBSCAN algorithms face a lack of fairness consideration, which may result in biased clustering outcomes. To solve this issue, this paper proposes an improved clustering algorithm—FairDBSCAN (FDBSCAN). Based on the traditional DBSCAN framework, FDBSCAN introduces the novel concept of Fair Directly Density-Reachable and incorporates an innovative dynamic fairness-aware mechanism. The algorithm leverages a dynamic adjustment strategy to balance clustering performance and fairness during the cluster expansion process. Additionally, FDBSCAN employs a probabilistic reassignment strategy to ensure that unassigned points are represented fairly. Experiments conducted on multiple datasets demonstrate that FDBSCAN significantly improves fairness while achieving clustering results comparable to those of traditional methods.