Component-wise design method of fuzzy C-means clustering validity function based on CRITIC combination weighting

作     者：Wang, Guan Wang, Jie-Sheng Wang, Hong-Yu Liu, Jia-Xu 

作者机构：Univ Sci & Technol Liaoning Sch Elect & Informat Engn Anshan 114044 Peoples R China 

出 版 物：《JOURNAL OF SUPERCOMPUTING》 (超高速计算杂志)

年 卷 期：2023年第79卷第13期

页      面：14571-14601页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Basic Scientific Research Project of Institution of Higher Learning of Liaoning Province [LJKZ0293] Postgraduate Education Reform Project of Liaoning Province [LNYJG2022137] 

主　　题：Fuzzy C-means clustering algorithm Clustering validity function CRITIC combination weighting Component-wise design 

摘      要：Clustering validity function is an important research direction in clustering problems. Its idea is to specify the number of data clusters in advance so as to judge the optimal partition result on data sets. Studies have shown that no clustering validity function can handle all types of data, and its performance is not consistently better than other indices. Therefore, a component-wise design method of the fuzzy C-means (FCM) clustering validity function based on CRITIC combination weighting is proposed by adopting components for evaluating clustering performance. The weighting method combines expert weighting and the coefficient of variation method (CRITIC), arranges and combines six validity components with weights, and generates 55 different fuzzy clustering validity functions. These clustering validity functions are then tested on six typical UCI data sets, and the function with the simplest structure and best classification performance is selected through comparison. Finally, it is compared with seven typical clustering validity functions and four common combination clustering validity evaluation methods on eight UCI data sets. The simulation results demonstrate that the proposed validity function can achieve better classification results and determine the optimal cluster number for different data sets.