基于邻域模型的K-means初始聚类中心选择算法

传统的K-means算法由于其方法简单，在模式识别和机器学习中被广泛讨论和应用。但由于K—means算法随机选择初始聚类中心，而初始聚类中心的选择对最终的聚类结果有着直接的影响，因此算法不能保证得到一个唯一的聚类结果。利用邻域模型中对象邻域的上下近似，定义了对象邻域耦合度和分离度的概念，给出了对象在初始聚类中心选择中的重要性，提出了一种初始聚类中心的选择算法。另外，分析了邻域模型中三种范数对聚类精度的影响，并和随机选择初始聚类中心、CCIA选择初始聚类中心算法进行了比较，实验结果表明，该算法是有效的。

Initial Cluster Centers Choice Algorithm for K-means Based on Neighborhood Model

The traditional K-means algorithm considered as a simple method has been widely discussed and applied in pattern recognition and machine learning.However,K-means algorithm can not guarantee unique clustering result because initial cluster centers are chosen randomly,moreover,choosing initial cluster centers is extremely important as it has a direct impact on the formation of final clusters.In this paper,concepts of coupling and division are defined by using low approximation and upper approximation of objec...