改进的数据挖掘模糊聚类算法研究与分析  

Study and Analysis on Improved Fuzzy Clustering Algorithm in Data Mining

作　　者：李艳 张庆 田苏慧敏 

机构地区：[1]宁夏师范学院物理与电子信息工程学院,宁夏固原756000

出　　处：《宁夏师范学院学报》2018年第1期36-47,共12页Journal of Ningxia Normal University

摘　　要：由于传统的模糊C均值(FCM)聚类算法具有不能自动确定聚类数目、收敛速度较慢、抗噪性差、对噪声数据比较敏感等诸多缺点,研究基于相对熵约束的FCM聚类算法,加入对手抑制式方法和基于分离度和紧致度的聚类函数,有效克服上述缺点.研究结果表明:改进后的FCM算法与传统的FCM聚类算法相比,收敛速度、准确度、精确度、特异度均有提高,且具有良好的抗噪性能,并能自动确定出最佳的聚类数目.The fuzzy C-means（FCM） clustering algorithm has the disadvantage that it is a slow convergence algorithm,and the number of clusters,which has the great uncertainty should be prior artificially given. It also has the disadvantage that the data is poor in anti-noise performance,and is sensitive to noise. In order to overcome these deficiencies of FCM algorithm,we add opponent suppression approach to accelerate the convergence rate based on the relative entropy constraint FCM clustering algorithm. In addition,we can calculate the quantity of clusters automatically by adding the cluster validity function. The results showed that the modified FCM algorithm enhances on the convergence speed,acuracy,precision and the specificity,and the anti-noise performance is better compared with the traditional FCM clustering algorithm,and the optimal number of clusters can be automatically determined.

关 键 词：数据挖掘 模糊C均值聚类 对手抑制式 聚类有效性函数 相对熵 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]