Abstract: Because of the problems of Kernel Fuzzy C-means Clustering algorithm（KFCM） easy falling into local optimality and the sensitivity to initial value, a kernel fuzzy C-means clustering based on Shuffled Frog Leaping Algorithm（SFLA） is presented. But its effect is not satisfactory for the data with larger clusters number and higher dimensions. So adaptive inertia weight is used to update the strategy of SFLA. Then the obtained optimal solution by improved shuffled frog leaping algorithm（ISFLA） is taken as initial clustering centers of KFCM algorithm to optimize initial clustering centers, so as to get the global optimum and overcome the shortcoming of the KFCM algorithm. The results of experiments on the artificial and real data show that compared with the KFCM and FCM clustering algorithm, the new algorithm optimization ability would be stronger, the number of iterations less, and the clustering effect better.

Key words: Kernel Fuzzy C-means Clustering（KFCM）, improved shuffled frog-leaping algorithm, cluster analysis, data mining

摘要： 针对核模糊C-均值（KFCM）聚类算法存在易陷入局部极小值，对初始值敏感的缺点。将混合蛙跳算法（shuffled frog leaping algorithm，SFLA）用于KFCM中，但在聚类数较大和维数较高时，聚类效果不理想，为此提出将自适应惯性权重引入混合蛙跳算法的更新策略中，再用改进后的混合蛙跳算法求得最优解作为KFCM算法的初始聚类中心，利用KFCM算法优化初始聚类中心，求得全局最优解，从而有效克服了KFCM算法的缺点。人造数据和经典数据集的实验结果表明，新算法与KFCM和FCM聚类算法相比，寻优能力更强，迭代次数更少，聚类效果更好。

关键词: 核模糊C-均值聚类, 改进的混合蛙跳算法, 聚类分析, 数据挖掘