多层递阶融合模糊特征映射的模糊C均值聚类算法

PDF下载HTML

Title:

Fuzzy C-means clustering algorithm for multilayered hierarchical fusion fuzzy feature mapping

作者:

鲍国强^1,2, 应文豪³, 蒋亦樟^1,2, 张英^1,2, 王骏^1,2, 王士同^1,2

1. 江南大学 数字媒体学院, 江苏 无锡 214122;
2. 江苏省媒体设计与软件技术重点实验室, 江苏 无锡 214122;
3. 常熟理工学院 计算机科学与工程学院, 江苏 常熟 215500

Author(s):

BAO Guoqiang^1,2, YING Wenhao³, JIANG Yizhang^1,2, ZHANG Ying^1,2, WANG Jun^1,2, WANG Shitong^1,2

1. School of Digital Media, Jiangnan University, Wuxi 214122, China;
2. Jiangsu Key Laboratory of Media Design and Software Technology, Wuxi 214122, China;
3. School of Computer Science and Engineering, Changshu Institute of Technology, Changshu 215500, China

关键词:

Takagi-Sugeno-Kang (TSK)模糊系统; 主成分分析(PCA); 无监督学习; 模糊C均值聚类

Keywords:

Takagi-Sugeno-Kang (TSK) fuzzy system; principal component analysis (PCA); unsupervised learning; fuzzy C-means clustering

分类号:

TP181

DOI:

10.11992/tis.201703047

摘要:

针对复杂非线性数据的无监督学习问题，提出一种新型的映射方式来有效提高算法对复杂非线性数据的学习能力。以TSK模糊系统的规则前件学习为基础，提出一种新型的模糊特征映射新方法。接着，针对映射之后的数据维度过大问题，引入多层递阶融合的概念，进一步提出基于多层递阶融合的模糊特征映射新方法，从而有效避免了因单层模糊特征映射之后特征维数过高而导致的数据混乱和冗余的问题。最后与模糊C均值算法相结合，提出基于多层递阶融合模糊特征映射的模糊C均值聚类算法。实验研究表明，文中算法相比于经典模糊聚类方法，有着更加优越、稳定的性能。

Abstract:

In this paper, we propose a novel feature mapping technique called multilayer hierarchical fusion fuzzy feature mapping for the unsupervised learning of complex nonlinear data and combine it with the classical fuzzy C-means clustering. Based on the regular antecedent learning of the Takagi-Sugeno-Kang (TSK) fuzzy system, we first propose a novel fuzzy feature mapping method. Then, to address big data dimensions by fuzzy feature mapping, we propose a fuzzy feature mapping mechanism based on multilayer hierarchical fusion. This mechanism combines fuzzy feature mapping with principal component analysis (PCA), thereby avoiding the data confusion and redundancy caused by the high dimensionality of single-layer fuzzy feature mapping. Finally, we develop a novel FCM clustering algorithm based on multilayered hierarchical fusion feature mapping. The experimental results show that, in comparison with classical fuzzy clustering methods, the performance of the proposed algorithm is superior and more stable.