摘要
针对传统的模糊核聚类算法(FKCM)需给出聚类个数,且对初始值敏感、易陷入局部最优的缺点,本文提出了一种基于高斯核化有效性指标的自适应优选聚类数的模糊核聚类算法(GKVI-AOCN-FKCM)。利用基于密度和距离的方法选取初始聚类中心,克服了对初始值的敏感,提高了聚类效率。然后用高斯核函数核化后的有效性指标评价聚类效果并自动确定最佳分类数,从而无监督地实现对数据集的模糊划分。对Iris数据集的仿真实验及石脑油属性数据分类的应用验证了算法的可行性和有效性。
To overcome the drawbacks of traditional fuzzy kernel clustering algorithm that clustering number need to be given in advance, being sensitive to the initial value and easy to be trapped into local optimum, the Adaptive Optimal Clustering Number of Fuzzy Kernel Clustering Method based on Gaussian Kernel Validity Index is proposed （GKVI-AOCN-FKCM）. The proposed algorithm uses density and distance-based method to select initial clustering center, and could solve the problem of sensitivity to initial value, with the efficiency of clustering improved. Then, the Gaussian kernel validity index is used to evaluate clustering effect and to determine the optimal number of categories. After that, the fuzzy partition of data set can be achieved unsupervised. The simulation in Iris data set and the application in naphtha attribute data classification verify the feasibility and effectiveness of the proposed method.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2012年第10期1199-1203,共5页
Computers and Applied Chemistry
基金
国家自然科学基金项目重点基金资助项(U1162202)
国家高技术研究发展计划(863)资助项目(2012AA040307)
上海市基础研究重点项目(10JC1403500)
上海市重点学科建设项目(B504)
流程工业综合自动化国家重点实验室开放课题基金资助
关键词
聚类分析
模糊聚类
高斯核函数
聚类中心初始化
有效性指标
cluster analysis, fuzzy clustering, Gaussian kernel function, initial cluster centers, validity index
