基于马氏距离的可能性熵聚类算法

Possible Entropy Clustering Algorithm Based on Mahalanobis Distance

针对模糊聚类基于欧氏距离准则没有考虑样本向量特征之间关联的问题,在可能性熵的基础上,提出了基于马氏距离的可能性熵聚类算法,考虑了熵聚类的全局效应和局部效应,算法具有清晰的物理意义和数字特征。同时,消除了样本之间的相关性与量纲差异造成的影响。在人工数据集和UCI数据集上,将所提算法与硬C均值聚类、模糊C均值聚类,可能性熵聚类进行比较,实验结果表明,上述方法可以有效地估算各类中心,同时,有效地减少了算法的迭代次数,提升聚类分析效率。相比于其它几种聚类算法,聚类精度有了一定提升。

In order to improve the problem of association between sample vector features based on the Euclidean distance criterion for fuzzy clustering,this paper considers the global and local effects of entropy clustering,and proposes the possibility entropy clustering based on Mahalanobis distance.The algorithm has a clear physical meaning and mathematical characteristics.At the same time,the influence of correlations and dimensional differences between samples can be eliminated.Experimental results show that PEC-M is more effective and feasible in classification than hard C-means clustering,fuzzy C-means clustering and possibility entropy clustering.At the same time,the iterative times of the algorithm are effectively reduced and the clustering analysis efficiency is improved.