基于双重迭代聚类的模糊投影寻踪聚类算法

A Dual Iterative Clustering Based Fuzzy Projection Pursuit Clustering Algorithm

  • 摘要: 建立了一种新的聚类算法——模糊投影寻踪聚类(fuzzy projection pursuit cluster, FPPC)算法,实现了投影寻踪聚类(projection pursuit clustering,PPC)算法与模糊聚类迭代(fuzzy clustering iterative,FCI)算法的良好融合。FPPC算法首先建立了一种新的投影指标函数,该函数由投影值标准差和投影点广义欧氏权距离平方和构成,能避免传统PPC中选取惟一参数密度窗宽时完全依赖经验来决定的问题;然后采用投影技术对高维数据进行降维处理,执行FCI步骤来对低维样本集进行初次聚类运算;接着通过寻找最优投影方向的过程,对样本集进行PPC的二重聚类。在FPPC求解过程中,运用了由混沌理论、文化算法与差分进化算法融合而成的混沌文化差分进化算法进行优化处理。实验仿真表明,FCI与PPC双重迭代聚类的FPPC算法拥有更优的聚类精度及有效性。

     

    Abstract: This paper presents a new fuzzy projection pursuit clustering (FPPC) algorithm. FPPC is a combination of the fuzzy clustering iteration (FCI) algorithm and the projection pursuit clustering algorithm. In this paper, we adopted a new projection index function formed by the standard deviation of projection values and the quadratic sum of Euclidean distance between projection values. The new projection index function can avoid the qualitative selection of the Density Window Width, which is generally determined by experience. After lowering the dimension of sample data using projection technology, the FPPC algorithm takes a dual iterative clustering approach with FCI and PPC. In the FPPC solution process, the chaotic culture differential evolution (CCDE) algorithm formed by the chaos theory, cultural algorithm and differential evolution algorithm is adopted. Experimental simulations show that FPPC algorithm has higher clustering precision and effectiveness.

     

/

返回文章
返回