当前位置: X-MOL 学术Soft Comput. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Kernel intuitionistic fuzzy c-means and state transition algorithm for clustering problem
Soft Computing ( IF 4.1 ) Pub Date : 2020-04-02 , DOI: 10.1007/s00500-020-04879-8
Xiaojun Zhou , Rundong Zhang , Xiangyue Wang , Tingwen Huang , Chunhua Yang

Abstract

Clustering problems widely exist in machine learning, pattern recognition, image analysis and information sciences, etc. Although many clustering algorithms have been proposed, it is unpractical to find a clustering algorithm suitable for all types of datasets. Fuzzy c-means (FCM) is one of the most frequently-used fuzzy clustering algorithm for the reason that it is efficient, straightforward, and easy to implement. However, the traditional FCM taking Euclidean distance as similarity measurement can not distinguish the intersection between two clusters. Therefore, kernel function has been taken as similarity measurement to solve this issue. As a comprehensive partition criterion, intuitionistic fuzzy set which consider both membership degree and non-membership degree has been used to replace traditional fuzzy set to describe the natural attributes of objective phenomena more delicately. Thus, Kernel intuitionistic fuzzy c-means (KIFCM) has been proposed in this paper to settle clustering problem. Considering FCM is easily getting trapped in local optima due to its high sensitivity to initial centroid. State Transition Algorithm (STA) has been adopted in this study to obtain the initial centroid to enhance its stability. The proposed STA-KIFCM compared with some other clustering algorithms are implemented using five benchmark datasets. Experimental results not only show that the proposed method is efficient and can reveal encouraging results, but also indicate that the proposed method can achieve high accuracy.



中文翻译:

聚类问题的核直觉模糊c均值和状态转移算法

摘要

聚类问题广泛存在于机器学习,模式识别,图像分析和信息科学等领域。尽管已经提出了许多聚类算法,但是找到适合于所有类型数据集的聚类算法是不切实际的。模糊c均值(FCM)是最常用的模糊聚类算法之一,因为它高效,直接且易于实现。然而,以欧几里德距离作为相似度度量的传统FCM无法区分两个聚类之间的交集。因此,内核函数已作为相似性度量来解决此问题。作为全面的划分标准,考虑隶属度和非隶属度的直觉模糊集已被用来代替传统的模糊集,以更精细地描述客观现象的自然属性。因此,本文提出了核直觉模糊c均值(KIFCM)来解决聚类问题。考虑到FCM对初始质心的高度敏感性,很容易陷入局部最优。本研究采用状态转换算法(STA)来获得初始质心以增强其稳定性。与其他一些聚类算法相比,所提出的STA-KIFCM是使用五个基准数据集实现的。实验结果不仅表明该方法是有效的并且可以显示令人鼓舞的结果,而且表明该方法可以达到较高的精度。

更新日期:2020-04-03
down
wechat
bug