基于PSO-TVAC的中心自适应权的FCM聚类算法  

FCM Clustering Algorithm Based on PSO-TVAC Algorithm with Adaptively Weighted Centers

作　　者：胡建华[1] 尹慧琳 

机构地区：[1]上海理工大学理学院,上海

出　　处：《应用数学进展》2021年第4期953-962,共10页Advances in Applied Mathematics

摘　　要：针对传统FCM算法依赖于初始聚类中心、对噪声敏感、容易陷入局部最优、分类时会倾向于多数类等缺点,本文首先提出一种基于PSO-TVAC的中心自适应权的FCM聚类算法(CWAFCM)。新算法将中心权重向量φ和自适应指数q引入目标函数,用以区分每个聚类中心的不同重要性;指数q和模糊因子m由粒子群算法(PSO-TVAC)优化;新提出一种聚类评价标准ACVI作为PSO-TVAC算法的适应度函数以提高聚类准确率。其次,将CWAFCM与过采样技术(SMOTE)相结合以适应于对不平衡数据聚类。通过对六个数据集(四个平衡数据集,两个不平衡数据集)进行仿真实验,结果表明CWFCM算法能够有效地优化聚类效果,且能提高不平衡数据集的聚类准确率。The traditional FCM algorithm relies on the initial clustering center, is sensitive to noise, is easy to fall into local optimum, and tends to classify most classes. In this paper, a FCM clustering algorithm based on PSO-TVAC algorithm with adaptively weighted centers is proposed. The new algorithm introduces the weight vector φ of centers and the adaptive exponent q into the objective function to distinguish the different importance of each cluster center. The exponent q and fuzzy factor m are optimized by particle swarm optimization (PSO-TVAC). Secondly, CWAFCM is combined with synthetic minority oversampling technique (SMOTE) to cluster unbalanced data. The results of experiments on six datasets (four balanced datasets and two unbalanced datasets) show that CWAFCM algorithm can effectively optimize the clustering effect and improve the clustering accuracy on unbalanced dataset.

关 键 词：模糊C均值算法 自适应权重 过采样技术 粒子群算法 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]