厦门大学学报（自然科学版）2026，Vol.65Issue(2)：340-348,9.DOI:10.6043/j.issn.0438-0479.202507011
基于二型模糊集数字特征的聚类方法及其应用
A clustering method based on numerical characteristics of type-2 fuzzy sets and its applications
摘要
Abstract
[Objective]In fuzzy clustering tasks,it is often difficult to predefine the fuzzifier parameter,and the defuzzification process of type-2 fuzzy sets imposes heavy computational overheads.These limitations restrict the efficiency and scalability of traditional interval type-2 fuzzy clustering models.To address these issues,we propose an efficient and robust type-2 fuzzy clustering approach that balances computational speeds and clustering accuracies.[Methods]A fast clustering algorithm based on numerical characteristics of type-2 fuzzy sets is developed.The proposed method integrates both the sample-to-cluster distance and these numerical characteristics of type-2 fuzzy memberships into the objective function,thereby achieving joint optimization of cluster centers and type-2 fuzzy memberships.During the iterative process,only these numerical characteristics are updated,and consequently the need for the computationally intensive Karnik-Mendel(KM)defuzzification procedure is eliminated.This simplification greatly reduces the time complexity and enhances the algorithm's robustness against noises.The theoretical framework of the algorithm establishes explicit update rules for cluster centers,membership centroids,and cardinalities,thus leading to convergence efficiency comparable to that of the classical fuzzy C-means(FCM)method.[Results]Experimental validation on two datasets demonstrates the effectiveness and stability of the proposed model.On the Wisconsin diagnostic breast cancer(WDBC)dataset,the proposed algorithm achieves a clustering accuracy of 72.84%under unsupervised and noisy conditions,successfully distinguishing real samples from artificial noises.On the IMDb movie review dataset,when the noise ratio increases from 0.1 to 1.0,the clustering accuracy of the proposed method rises slightly from 71.67%to 72.05%.This result demonstrates that the proposed method outperforms the second-best method(FCM,6 9.6 5%).The average running time increases modestly from 2.18 s to 3.63 s,which is only marginally slower than FCM(2.37 s)but far faster than the interval type-2 FCM(24.43 s).These results indicate that the algorithm maintains high accuracy and robustness with nearly linear computational scalability.[Conclusions]The proposed characteristic-based type-2 fuzzy clustering framework effectively mitigates the computational complexity and parameter sensitivity inherent in traditional type-2 fuzzy clustering methods.By leveraging interpretable numerical descriptors,centroid and cardinality,it retains the uncertainty modeling capacity of type-2 fuzzy sets and at the same time achieves high clustering accuracy,strong noise tolerance,and low computational cost.关键词
二型模糊聚类/数字特征/模糊聚类/电影类簇Key words
type-2 fuzzy clustering/numerical features/fuzzy clustering/movie classification分类
信息技术与安全科学引用本文复制引用
李志伟,张荣宇,杨昔阳..基于二型模糊集数字特征的聚类方法及其应用[J].厦门大学学报（自然科学版）,2026,65(2):340-348,9.基金项目
福建省自然科学基金项目(2025J01965) （2025J01965）
泉州市科学技术局2025年第一批高层次人才创新创业项目(2025QZC09R) （2025QZC09R）
