基于核方法的分类型属性数据集模糊聚类算法

针对分类型属性数据的聚类问题．将核方法的思想推广到快速、高效率的模糊c-均值算法，构造了基于核函数的模糊核c-均值聚类算法．该算法通过使用经验核矩阵充分利用了数据间的“相异性”信息，并且避免了模糊k-modes算法中每次迭代均要直接计算类中心的缺点，提高了聚类的精确度和稳定性．同时该算法对模式(类中心)的初始值选择不敏感．时实际的线性可分的和线性不可分的分类型属性数据集的仿真实验证明了该算法的有效性．

A New Fuzzy Clustering Algorithm of Categorical Data Set Based on the Kernel Method

Aiming at the clustering of the categorical data a nd by extending the kernel method to the fast efficient fuzzy c -means clustering algorithm, a fuzzy kernel c -means (FKCM) clustering algorithm based on the kernel function was constructed. In this algorithm, the empirical kernel matrix is applied in order to fully utilize the dissimilarity information among the data. Unlike the fuzzy k -mode algorithm, the modes in the new algorithm need not be directly calculated in each iteration, thus improving the precision and stability of the clustering algorithm and making the new algorithm insensitive to the selection of mode(centroid) initialization. A simulation was finally carried out on the actual linearly and nonlinearly separable categorical data sets, which demonstrates the effectiveness of the proposed algorithm.