基于减法聚类的GK模糊聚类研究

摘要

Gustafson-Kessel (GK) algorithm is one of the most widely used fuzzy clustering algorithms. But this algorithm is hypersensitive to the setting of the initial value and is easy to fall into local optimal solution. Whafs more,GK algorithm requires a given clustering number. It is poor at self-adjustment. In view of the above shortcomings of GK algorithm, we adopt the subtractive clustering algorithm to initialize the GK algorithm, which can reflect the data structure better. Based on the initial value applied by the subtractive clustering algorithm,we adopt the clustering validity function to determine the reasonable clustering number in order to achieve the automatic classification and reasonable clustering partitioning results. Finally, the simulation experiments to the artificial data set and the iris data set demonstrated that the improved algorithm can automatically determine the reasonable clustering number and the clustering correctness increased obviously.%Gustafson-Kessel(GK)算法是目前应用最广泛的模糊聚类算法之一.但是它对初值的设置非常敏感,容易陷入局部最优解；该算法还必须事先给定聚类个数,自我调节能力差.针对GK算法上述缺点,采用减法聚类对GK聚类算法进行初始化,初值设置更能反映数据集结构；基于减法聚类提供的初值,采用聚类有效性函数确定合理的聚类类别数,以达到自动分类的效果能给出较为合理的聚类划分结果.通过对人工数据集和iris数据集的仿真实验,表明改进算法在自动确定合理聚类类别数的同时,聚类正确率明显提高.