基于区域比例的聚类方法

为了改善DBSCAN参数敏感性和对密度分布不均数据对象聚类质量不高的问题，提出了一种基于DBSCAN算法的改进聚类方法。算法使用K最近邻的均值距离度量密度，中心点选取当前密度最大点，并以中心点为核心点扩展种子队列，直至由给定的密度比例因子所决定的密度边缘。为了改善聚类质量，提出了候选核心点，并使用给定的半径比例因子发现核心点。在实验中，利用数据集对该算法进行了测试，测试结果证明了该改进算法的参数鲁棒性，和在聚类密度分布不均数据集时的较好性能。

Clustering algorithm based on local scaling

In order to improve the robustness of parameters and undergrade clustering quality when clustering dataset with maldistribution of density,the paper presents an improved clustering algorithm based on DBSCAN.The algorithm uses average distance of knn objects to measure the density of each object.The center object is defined as the object of local maximal density,the seed expands from center point until the edge of density which is defined by scale-factor of density.Candidate core object is used to improve the quality of clustering by searching core object with scale-factor of radius.In experiment,datasets are used to test the algorithm,which proves the robustness of parameters of the improved algorithm and excellent performance when clustering dataset with maldistribution of density.