Abstract: In order to cluster multi-density dataset, a clustering algorithm based on density-reachable for multi-density is proposed. Grid partition method is used to improve efficiency when computing each point’s density. A clustering starts with the highest density point and uses expansion to form a cluster based on density-reachable and breadth-first strategy. Experimental results show that this algorithm can effectively discover clusters of arbitrary shapes for multi-density and uniformity density data sets with noises. It can get good cluster quality and is more efficient than SNN algorithm.

Key words: clustering algorithm, neighborhood grid, density-reachable, breadth-first, multi-density

摘要： 为对多密度数据集聚类，提出一种基于密度可达的多密度聚类算法。使用网格划分技术来提高计算每个点密度值的效率，每次聚类都是从最高密度点开始，根据密度可达的概念和广度优先的策略逐步向外扩展进行聚类。实验表明，该算法能够有效地对任意形状、大小的均匀数据集和多密度数据集进行聚类，并能较好地识别出孤立点和噪声，其精度和效率优于SNN算法。

关键词: 聚类算法, 邻域网格, 密度可达, 广度优先, 多密度

CLC Number: 

• TP301.6