摘要
随着聚类算法在智能测量系统、多源传感数据分析与嵌入式状态识别等场景中的广泛应用,如何在保证聚类质量的同时兼顾敏感属性的公平性,已成为制约聚类算法在关键测量任务中应用效果的瓶颈问题。为解决上述问题,提出了一种创新的种群优化联合鲁棒距离度量的公平性K-means聚类算法(PODM-Kmeans)。该方法在构建过程中,充分考虑到敏感属性的公平性与聚类质量之间的平衡性,引入改进的布谷鸟搜索算法以实现初始聚类中心选择过程中的全局搜索能力和局部搜索能力的平衡,有效增强了聚类效果的稳定性。在此基础上,在聚类迭代目标函数的构建上,该方法有效采用了公平性约束和簇大小约束机制,并融合了灵活的加权欧氏范数作为距离度量方法,合理抑制了异常值所带来的消极影响,助力了公平性的提升。通过在5个合成数据集和5个真实数据集上进行的大量实验结果表明,PODM-Kmeans在同类方法中具有较优的性能表现,尤其在Adult、Bank、Census1990和CreditCard 4个数据集上,在维持一定的聚类效果的同时,PODM-Kmeans的公平性比率(FR)指标均超过0.95。
With the widespread application of clustering algorithms in intelligent measurement systems,multi-source sensor data analysis,and embedded state recognition,ensuring fairness with respect to sensitive attributes while maintaining clustering quality has become a key challenge that limits their effectiveness in critical measurement tasks.To address this issue,we propose a population optimization combined with robust distance metric for fair K-means clustering method(PODM-Kmeans).The proposed method balances clustering quality and fairness by incorporating an enhanced Cuckoo Search algorithm to achieve a trade-off between global and local search capabilities during the initialization of cluster centers,thereby improving clustering stability.Furthermore,fairness constraints and cluster size constraints are effectively integrated into the iterative clustering objective function.A flexible weighted Euclidean norm is adopted as the distance metric to mitigate the negative impact of outliers,contributing to improved fairness.Extensive experiments conducted on five synthetic and five real-world datasets demonstrate the superior performance of PODM-Kmeans compared to existing methods.Notably,on the Adult,Bank,Census1990,and CreditCard datasets,PODM-Kmeans achieves a fairness ratio(FR)exceeding 0.95 while maintaining high clustering quality.
作者
谢一涵
毕鹏飞
王爱萍
Xie Yihan;Bi Pengfei;Wang Aiping(School of Artificial Intelligence(School of Future Technologies),Nanjing University of Information Science&Technology,Nanjing 210044,China;School of Software,Nanjing University of Information Science&Technology,Nanjing 210044,China)
出处
《电子测量与仪器学报》
北大核心
2025年第6期121-133,共13页
Journal of Electronic Measurement and Instrumentation
基金
江苏省基础研究计划基金(BK20220452)项目资助。
关键词
K-MEANS聚类
公平性
种群优化
鲁棒距离度量
布谷鸟搜索算法
欧式距离
K-means flustering
fairness
population-based optimization
robust distance metric
cuckoo search algorithm
Euclidean distance
作者简介
谢一涵,现为南京信息工程大学人工智能学院(未来技术学院)本科生,主要研究方向为机器学习、聚类分析、深度学习、目标检测。E-mail:202383290356@nuist.edu.cn;通信作者:毕鹏飞,2021年于哈尔滨工程大学获得博士学位,现为南京信息工程大学讲师,主要研究方向为机器学习、深度学习、模式识别、水下目标感知。E-mail:pfcx@nuist.edu.cn;王爱萍,现为南京信息工程大学软件学院本科生,主要研究方向为边缘计算、数据分析。E-mail:1816463803@qq.com。
