一种改进的快速聚类算法GLDBSCAN

本文在基于局部密度的空间聚类算法LDBSCAN的基础上提出了一种基于网格和SP-Tree的快速聚类算法GLDBSCAN。改进算法设计了一种新的对数据空间进行划分的网格划分方法,并采用空间索引SP-Tree来组织网格结构。算法同时提出用网格中心对象来代表网格包含的对象集合进行聚类,从而降低聚类时间和I/O消耗,实现快速聚类。实验结果表明,GLDBSCAN算法在不降低LDBSCAN聚类质量的前提下,大大提高了聚类的效率,并且能有效地对大规模数据集进行聚类。

GLDBSCAN: An Improved Fast Clustering Algorithm

A fast algorithm GLDBSCAN is proposed in this paper,which considerably speeds up the original LDBSCAN.The new algorithm presents a novel approach to grid generation and adopts the spatial index structure SP-Tree to organize the grids.In addition,GLDBSCAN chooses the representative object in a grid to represent all objects of the grid to cluster,which reduces the execution frequency and I/O cost dramatically.The experimental results show that GLDBSCAN is much more efficient than LDBSCAN without reducing the quality of clustering results and it is effective and efficient in clustering large-scale datasets.