新的基于网格的聚类算法(CABG)利用网格处理技术对数据进行了预处理,能根据数据分布情况动态计算每个单元格的半径,并成功地将网格预处理后所得单元格数据运用于其后的聚类分析中,从而简化了算法所需的初始参数。实验表明,CABG算法不仅具有DBSCAN算法准确挖掘各种形状的聚类和很好的噪声处理能力的优点,而且具有较高聚类速度以及对初始参数较低的敏感度。
This paper presented a new grid-based clustering algorithm which preprocessed the data using grid processing method, The algorithm was capable of computing the radius of each grid according to the density dynamically. The required initial parameters for the clustering analysis were simplified by the previously processed data. The result of the experiments demonstrate that CABG is as accurate in discovering density-changeable clustering and handling of noise as DBSCAN, but CABG has higher clustering speed and less sensitivity to the initial parameters.