王志豪1,,,
潘金艳2,
罗斯哲1,
王德鑫1
1.厦门大学航空航天学院 厦门 361102
2.集美大学信息工程学院 厦门 361021
基金项目:国家自然科学基金(61203176)，福建省自然科学基金(2013J05098, 2016J01756)
详细信息
作者简介:高云龙：男，1979年生，副教授，主要研究方向为机器学习、时间序列分析和生产制造系统优化和调度
王志豪：男，1993年生，硕士生，研究方向为机器学习和模式识别
潘金艳：女，1978年生，副教授，主要研究方向为人工智能和机器学习理论与方法
罗斯哲：男，1995年生，硕士生，研究方向为模式识别和维数约简
通讯作者:王志豪 zhwang@stu.xmu.edu.cn
中图分类号:TP391; TP273计量
文章访问数:1859
HTML全文浏览量:1390
PDF下载量:94
被引次数:0
出版历程
收稿日期:2019-07-24
修回日期:2020-03-13
网络出版日期:2020-04-09
刊出日期:2020-07-23
Robust Fuzzy C-Means Based on Adaptive Relaxation
Yunlong GAO1,Zhihao WANG1,,,
Jinyan PAN2,
Sizhe LUO1,
Dexin WANG1
1. College of Aeronautics and Astronautics, Xiamen University, Xiamen 361102, China
2. College of Information Engineering, Jimei University, Xiamen 361021, China
Funds:The National Natural Science Foundation of China (61203176), The Provincial Natural Science Foundation of Fujian Province (2013J05098, 2016J01756)
摘要
摘要:噪声是影响聚类结果的最重要的因素之一，现有的模糊聚类算法主要通过对隶属度约束进行松弛的方式来降低噪声样本的影响。这种方式仍然存在两个基本问题需要解决：第一，如何评估一个样本是噪声的可能性；第二，如何在抑制噪声样本影响力的同时，保留正常样本的作用力。针对这两问题，该文提出了基于自适应松弛的鲁棒模糊C均值聚类算法(AR-RFCM)。新模型基于K最近邻的方式(KNN)来估计样本的可靠性，自适应地调整松弛参数，从而实现在降低噪声样本影响力的同时，保留可靠样本的作用力。此外，AR-RFCM利用了C均值聚类模型中隶属度的稀疏性来提高可靠样本的作用力，从而提高数据簇的内聚程度，进而降低噪声样本的影响。实验表明，AR-RFCM不仅在处理噪声样本时具有良好的鲁棒性，同时在25个UCI 数据集实验中，分类正确率(兰德指数)平均高于FCM算法7.7864%。
关键词:噪声/
聚类/
模糊C均值/
自适应/
松弛
Abstract:Noise is one of the most important influences for clustering. Existing fuzzy clustering methods try to reduce the impact of noise by relaxing the constraint condition of membership. But there are still two basic problems to be solved. The first is how to evaluate the probability that a sample point is a noise. The second is how to retain the effect of normal points while suppressing the impact of noise. To solve these two problems, Robust Fuzzy C-Means based on Adaptive Relaxation (AR-RFCM) is proposed. The new model estimates the reliability of sample points by the method of the K-Nearest Neighbor (KNN). It adjusts adaptively the relaxation parameters to reduce the impact of noise, and keeps the effect of reliable sample points at the same time. In addition, AR-RFCM utilizes the sparsity of membership in K-means to improve the effect of reliable sample points. Therefore, the compactness of clusters is improved and the impact of noise is suppressed. Experiments demonstrate that AR-RFCM has a good robustness for noise, and also achieves higher rand index in all 25 UCI data sets, even averagely higher than FCM 7.7864%.
Key words:Noise/
Clustering/
Fuzzy C-Means (FCM)/
Adaptive/
Relaxation
PDF全文下载地址:
https://jeit.ac.cn/article/exportPdf?id=e70a1c3a-9e53-410c-9aec-a5b4fa9f0df5
