一种改进的K-Means算法

An Improved K-Means Clustering Algorithm

传统K-Means对算法使用者有较高的要求,需要明确K值,并确定初始中心点的位置。通过定义、检测并删除离群点,运用Canopy算法辅助确认K值范围和粗略中心点,借助Silhouette评价指标选择最优K值及其对应的聚类结果的方法,对传统K-Means算法进行改进,改进后的算法不需要手工输入K值和初始中心点。验证结果表明:改进的K-Means算法在聚类时,结果稳定准确,且当数据点数量较大时在迭代次数方面略优于传统算法。

The traditional K-Means has a high requirement for the user of the algorithm,need determine the K value and the location of the initial center point, define, detect and delete the outliers, canopy algorithm assists in identifying the range of K values and rough center points; select the optimal K value for Silhouette evaluation index and its corresponding clustering results, improve the traditional algorithm of K-means, the improved algorithm does not need to manually enter the K value and the initial center point.The results show that the improved k-means algorithm is stable and accurate when it is clustered; and when the number of data points is large, the number of iterations is slightly better than the traditional algorithm.