一种有效的K-means聚类中心初始化方法*

传统K-means算法由于随机选取初始聚类中心，使得聚类结果波动性大；已有的最大最小距离法选取初始聚类中心过于稠密，容易造成聚类冲突现象。针对以上问题，对最大最小距离法进行了改进，提出了最大距离积法。该方法在基于密度概念的基础上，选取到所有已初始化聚类中心距离乘积最大的高密度点作为当前聚类中心。理论分析与对比实验结果表明，此方法相对于传统K-means 算法和最大最小距离法有更快的收敛速度、更高的准确率和更强的稳定性。

Effective method for cluster centers' initialization in K-means clustering

Initializing cluster centers randomly, traditional K-means algorithm leads to great fluctuations in the clustering results. The existing max-min distance algorithm, indeed, has rather dense cluster centers, which may easily bring about clustering conflicts. To solve these problems, this paper regarded the existing max-min distance algorithm as the thinking foundation and proposed the maximum distances product algorithm. Based on the theory of density-based clustering, the maximum distances product algorithm selected each point which had maximum product of distances between itself and all other initialized clustering centers. Theory analysis and experimental results show that compared with traditional K-means algorithm and max-min distance algorithm, the maximum distances product algorithm can result in faster convergence speed, higher accuracy, greater stability.