基于核的可能性聚类算法

针对模糊C-均值算法聚类分析时的缺陷，采用能够较好地处理噪音和孤立点的可能性聚类算法，并将核学习方法的思想应用于可能性聚类算法中，提出一种基于核的可能性聚类算法。该方法利用Mercer核将观察空间的待分类样本点经过一个非线性映射后，映射到一个高维的核空间，突出不同类别样本之间的特征差异，使得原来线性不可分的样本点在核空间中变得更加线性可分，从而更好地聚类。经仿真实验表明，基于核的可能性聚类算法比模糊C-均值以及可能性聚类算法具有更好的聚类效果，且算法能够很快地收敛。

Kernel-based possibilistic clustering algorithm

Aimed at the default of fuzzy C-means which is highly sensitive to noise data and outliers and is confined to the distribution of class,possibilistic clustering algorithm is adopted which can preferably deal with noise data and outliers.The idea of kernel-based learning method is applied to possibilistic clustering algorithm and a kernel-based possibilistic clustering algorithm is presented.By using Mercer kernel function,the data is mapped from original space to high-dimensional kernel space where feature differences among all kinds of samples are stood out and the data are expected to more separable in order that clustering is better performed.The simulation re-sults show that the kernel-based possibilistic clustering algorithm achieve better clustering effect than fuzzy C-means and possibilistic clus-tering algorithm,and is fast in convergence.