基于数据分段的K-means的优化研究

Optimizing Research on K-means Based on Data Partition

K-means聚类算法是一种主流的迭代下降聚类算法,收敛于局部最优化状态。由于K-means随机选取k个初始聚类中心,使得聚类结果的有效性随初始输入而波动,为此文中采取一种预处理的方式来选取初始聚类中心。首先在某种范数的意义下,确定相隔最远的两个数据点之间的距离,然后采用数据分段的方法,将数据集分成k段,在每段中选取一个中心,以此来减小聚类结果随初始输入的波动。实验显示优化后的K-means有效地消除了初始输入的影响,并显著地减少了算法迭代次数和聚类误差。

The K-means clustering algorithm is one kind of mainstream iterative drop clustering algorithm,which restrains in the partial optimized state.Because K-means randomly selects initial clustering center,which result in the result of clustering is obviously fluctuate along with the initial input.Thus this paper adopts the pretreatment way to select the initial clustering center.First under one kind of norm,calculate out the farthest distance,then use the method of data partition to divide the data set into k section and select a center in each section.The experiment demonstrates the optimizing K-means eliminate the initial input influence,effectively reduced the iteration number of times and clustering error.