基于模糊交叉网格的初始聚类中心选取方法

Algorithm for Initialization of Clustering Center Based on Fuzzy Crossing Grid

选取初始聚类中心是多数聚类算法的首要步骤,往往影响着聚类的效果。为了避免算法迭代过程中易陷入局部最优的问题,本文提出了一种基于模糊交叉网格的初始聚类中心选取方法。算法通过对数据空间网格化后,以网格交点为中心的邻近网格组成网格空间,根据数据点的隶属度统计每个网格空间的密度,再通过局部最大网格空间选取K个初始聚类中心。在真实数据集上进行实验,结果表明该方法在保证了聚类效果的同时,提高了收敛速度。

Algorithm for initialization of clustering center is the preliminary step for the clustering algorithm which influences the effect of clustering. Thus the algorithm for initialization of clustering center based on fuzzy crossing grid is proposed to avoid the local optimization in the process of iterative algorithm. This new algorithm constructs the grid center on the basis of the neighboring grids taking the crossing point of the grids as center through the meshing of data space and calculates the density of each grid according to the membership grade of the data points and then selects K initialization of clustering centers. Results of the experiments based on the real data sets demonstrate that this algorithm ensures the clustering effect and accelerates the convergence as well.