融合K-means和CFSFDP的聚类算法

A Clustering Method Based on K-means and Fast Density Peak Search Algorithm

在K-means算法中,初始k值和初始聚类中心对聚类结果都有影响。针对K-means算法存在的问题,提出一种K-means算法与聚类的快速搜索和发现密度峰算法结合的聚类算法(K-CFSFDP)。该算法思想:先采用CFSFDP算法得到每个数据点的ρ_i和δ_i并将其作为数据新的特征向量,再次使用CFSFDP算法,对新的ρ_i进行升序排列,以斜率的变化自动选择聚类中心点;再运用K-means算法进行迭代聚类。该算法在UCI数据集上能够又好又快又稳定的聚类。

In the K-means algorithm, the initial k-value and the initial cluster center have an effect on the clustering results. Aiming at the problems of K-means algorithm, a clustering algorithm(K-CFSFDP) combining K-means algorithm with clustering fast search and discovery density peak algorithm is proposed. The idea of the algorithm: Firstly, the CFSFDP algorithm is used to obtainρi andδi of each data point and use it as the new feature vector of the data. The CFSFDP algorithm is used again to sort the newρi in ascending order, and the cluster center point is automatically selected with the change of the slope. Then use the K-means algorithm for iterative clustering. The algorithm is able to cluster well, fast and stable on the UCI dataset.