K-means initial clustering center optimal algorithm based on estimating density and refining initial

摘要

The performance of K-means clustering algorithm strongly depends on the initial parameters. Based on the segmenting algorithm of density estimation and large scale data group segmenting algorithm of the initial value limitation, a new algorithm for initializing the cluster center is presented. The idea of segmenting base on density is combined with the idea of sampling and the new idea is presented. The accuracy of sampling is improved by averagely segmenting every dimension of the database. The speediness of the refining initial algorithm ensures the new algorithm has superiority on time. The experiment demonstrates that the new algorithm has superiority on time and accuracy with other algorithms.