KNN优化的密度峰值聚类算法

KNN Optimized Density Peak Clustering Algorithm

针对K-means聚类算法存在初始聚类中心敏感,离群点的检测和去除困难的问题,提出基于K-最近邻(K-Nearest Neighbor,KNN)优化密度峰值的K-means聚类算法(K-Nearest Neighbor of Density Peaks Clustering-K-Means,KDPC-K)。首先利用KNN最近邻思想结合改进的密度函数确定数据样本点局部密度,其次以平均局部密度作为阈值来检测和去除离群点,最后采用一种自适应策略合并相似的聚类簇,获取初始聚类中心。实验在多个UCI数据集和含有噪声的人工数据集上表明,改进的算法与文献中的算法对比,具有更好的抗噪性、鲁棒性和聚类效果。

Aiming at the problem that the K-means clustering algorithm is sensitive to the initial clustering center and difficult to detect and remove outliers,a K-means clustering algorithm based on K-nearest neighbor(KNN)optimized density peak(KDPC-K)is proposed.First,the KNN nearest neighbor idea combined with the improved density function is used to determine the local density of data sample points.Then,the average local density is used as a threshold to detect and remove outliers.Finally,an adaptive strategy is adopted to merge similar clusters to obtain the initial cluster centers.Experiments on multiple UCI data sets and artificial data sets containing noise indicate that the improved algorithm has better noise resistance,robustness and clustering effect compared with the algorithm in the literature.