一种融合 K-means 和快速密度峰值搜索算法的聚类方法

摘要

K-means 算法的初始聚类中心是随机选取的，不同的初始中心输入会得出不同的聚类结果。针对 K-means 算法存在的问题，提出一种融合 K-means 算法与聚类的快速搜索和发现密度峰算法的聚类算法（K-CBFSAFODP）。该算法是这样考虑的：类簇中心被具有较低局部密度的邻居点包围，且与具有更高密度的任何点都有相对较大的距离，以此来刻画聚类中心；再运用 K-means算法进行迭代聚类，弥补了 K-means 聚类中心随机选取导致容易陷入局部最优的缺点；并且引入了熵值法用来计算距离，从而实现优化聚类。在 UCI 数据集和人工模拟数据集上的实验表明，融合算法不仅能得到较好的聚类结果，而且聚类很稳定，同时也有较快的收敛速度，证实了该融合算法的可行性。%The initial clustering centre of K-means algorithm is selected randomly,different initial centre inputs will get different clustering results.Aiming at this problem of K-means algorithm,we proposed a clustering algorithm which combines K-means algorithm and clustering with the fast density peaks search and finding algorithm (K-CBFSAFODP).This algorithm has the following considerations:the class cluster centre is surrounded by neighbour points with lower local density,and has relatively larger distance to any point with higher density,this is used to depict the cluster centre;then the K-means algorithm is employed for iterative clustering,this makes up the defect that to randomly select K-means clustering centre leads to falling into local optima easily.Moreover,the algorithm introduces entropy method to calculate the distance,thereby realises the optimisation of clustering.It is demonstrated by the experiments on UCI datasets and artificial simulation dataset that this combination algorithm can get better clustering results,and the clusters is very stable as well;meanwhile it also has fast convergence speed.These confirm the feasibility of the combination algorithm.