【摘要】 提出了一种基于网格距离的融合式聚类算法(Agglomerative Clustering algorithm based on Grid Distance,ACGD)。为规模不同的数据集分别设计了初始球状网格和初始矩形网格,并以此作为合并过程的起点。基于随机映射思想设计了网格之间的距离定义并以此完成聚类任务。ACGD的参数以自适应学习策略确定。真实数据集上的实验表明,ACGD具有良好聚类效果,具有比同类算法更高的效率和算法鲁棒性。更多还原

【Abstract】 Proposeed an agglomerative clustering algorithm based on grid distance,named as ACGD.ACGD starts initial aggregation from data grids instead of data points.For large-sized dataset and small-sized dataset,sphere-shaped grid and rectangular-shaped initial grid were designed respectively.The aggregating process was conducted based on a novel grid distance definition,which is developed according to random projection idea.ACGD is equipped with the self-tuning parameterization strategies.Experimental evidence of real datasets demonstrates the fine clustering performance of ACGD,and its advantage in efficiency and robustness over its peers.更多还原