优化核参数的模糊C均值聚类算法

Kernel-based fuzzy C-means clustering method based on parameter optimization

核模糊C均值聚类算法(Kernel-based fuzzy C-means clustering method,KFCM)的性能受核参数的影响很大,然而实践中核参数的选择是极其困难的。为了解决这个问题,本文基于样本在高维空间中的类内距离近、而类间距离远这一思路,提出了一种优化核参数的模糊C均值算法(Parameter optimation-based KFCM,POKFCM)。该算法首先利用K均值方法对样本集进行初始聚类,再通过比较实际核函数矩阵与理想核函数矩阵的相似性距离来确定最优核参数,最后将优化的核参数应用于核模糊C均值聚类算法。在6组UCI数据集上进行对比实验,结果表明POKFCM能有效地改善KFCM的聚类性能。

Kernel-based Fuzzy C-means Clustering Method（KFCM）is a common method for data clustering.The performance of KFCM is greatly affected by the parameter of the kernel function,while the selection of kernel parameter is extremely difficult in practice.To solve this problem,a Parameter Optimization-based KFCM（POKFCM）is proposed according to the idea that the distances between samples of the same class are closer than the distance between samples from different classes.First,initial clustering of dataset is completed by K-means method.Then the optimal kernel parameter is determined by calculating the distance similarity between the actual kernel matrix and ideal kernel matrix.Finally,the optimal kernel parameter is applied to KFCM.Clustering experiment results of six UCI datasets illustrate that POKFCM can effectively improve the clustering performance of KFCM.