基于类间差异最大化的加权距离改进K-means算法

为了改善K-means算法的聚类效果，将聚类准则函数定义为加权的类内误差平方总和SSE(sum of the squared error),并调整了K-means算法迭代过程中重新分配数据对象的方法:使用一个带有类内数据对象数的加权距离作为重新分配数据对象的依据,同时按类间差异最大化为准则优化了加权距离中的参数。实验表明,改进后的K-means算法可以在很大程度上减少大类被拆分情况的发生,明显改善聚类效果。

An improved K-means algorithm by weighted distance based on maximum between-cluster variation

To find natural clusters, the criterion function was improved by being defined as the weighted sum of the squared error. The way each point being assigned to the centroid in the iteration of the K-means algorithm was also modified: each point was assigned to the centroid that had minimum weighted distance. The weight was related with the number of points in each cluster, and the parameter of weighted distance was optimized by maximizing the between-cluster variation. Experimental results showed that the improved K-means algorithm significantly enhanced the clustering quality by reducing the probability of larger cluster’s being broken.