部分监督加权模糊C-均值算法的聚类分析

Clustering Analysis of Partial Supervised and Weighted Fuzzy C-Means Algorithm

模糊C-均值(FCM)算法具有对样本集进行等划分趋势的缺陷,对于团状、每类样本数相差较大的数据集, FCM算法的最优解可能不是数据集的正确划分,基于以上原因,以少量的先验知识作为部分监督信息,再利用样本点分布密度大小作为权值,提出了一种新的部分监督加权模糊C-均值(PSWFCM)算法,并且该算法的加权系数的计算和点密度范围限定值的选取都具有客观性。仿真结果证明,PSWFCM算法不仅在一定程度上克服了FCM算法的缺陷,而且具有良好的收敛性和鲁棒性,聚类效果也有较好的改善。

Based on fuzzy C-means (FCM) algorithm having limitation of equal partition trend for data sets, optimum clustering result of FCM algorithm might be not valid for demarcation of data sets haivng mass shape and large discrepancy of every class specimen number. A new partial supervised and weighted fuzzy C-means (PSWFCM) algorithm has been proposed in view of above-mentioned reasons, in which a little known knowledge is regarded as partial supervised information, distributing density size of data dot is regarded as weighted value, and calculation of weighted coefficient and choice of spot density range restriction value are objective. The simulation result proves that the algorithm has not only to certain extent overcome the limitation of FCM algorithm, but also has favorable convergence and robustness, and the clustering effect has been obviously improved.