分段二次方转换函数驱动的高斯核模糊C均值聚类  

Gaussian kernel fuzzy C-means clustering driven by piecewise quadratic transfer function

作　　者：唐益明[1,2] 张征 芦启明 TANG Yi-ming;ZHANG Zheng;LU Qi-ming(School of Computer and Information,Hefei University of Technology,Hefei 230601,Anhui,China;Anhui Province Key Laboratory of Affective Computing and Advanced Intelligent Machine,Hefei University of Technology,Hefei 230601,Anhui,China)

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230601 [2]合肥工业大学情感计算与先进智能机器安徽省重点实验室,安徽合肥230601

出　　处：《山东大学学报（理学版）》2020年第3期107-112,120,共7页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(61673156,61432004,U1613217,61672202);中国博士后科学基金资助项目(2014T70585);安徽省自然科学基金资助项目(1408085MKL15,1508085QF129);国家“863”高技术研究发展计划资助项目(2012AA011103)。

摘　　要：基于转换数据的模糊聚类算法存在转换模式单一、聚散效果不明显的问题,提出了分段二次方转换函数驱动的高斯核模糊C均值聚类算法。首先,通过分段二次方转换函数将原先分段线性数据转化的策略进行了相应的拓展,使数据转化的模式更加细腻,使得同类型的数据更好地聚集在一起,非同类型的数据进行远离。其次,我们引入了高斯核函数,将数据从低维空间映射到高维空间来进行聚类划分。最后,将这些集成到模糊聚类的框架之中,形成了所提算法。通过对比实验表明,所提算法明显优于相关的4种算法。As for the fuzzy C-Means algorithm based on transformed data, whose conversion mode is single while the effect of non-distribution is not obvious. For these reasons, the Gaussian kernel fuzzy C-means algorithm driven by piecewise quadratic transfer function is proposed. Firstly, the strategy of transforming the original piecewise linear data is extended by the piecewise quadratic conversion function, which makes the data conversion mode more delicate, so that the same type of data is better gathered together, and different types of data keep away. Secondly, we introduce a Gaussian kernel function to map data from low-dimensional space to high-dimensional space for clustering. Finally, the piecewise quadratic conversion function and Gaussian kernel function are integrated into the framework of fuzzy clustering to form the proposed algorithm. The comparison experiments show that the proposed algorithm is significantly better than the related four algorithms.

关 键 词：模糊C均值聚类算法 高斯核函数 非线性数据转换 二次方函数 

分 类 号：O159[理学—数学] TP18[理学—基础数学]