【摘要】 DPC算法是一种能够自动确定类簇数和类簇中心的新型密度聚类算法,但在样本分配策略上存在聚类质量不稳定的缺陷.其改进算法KNN-DPC虽然具有较好的聚类效果,但效率不高而影响实用.针对以上问题,文中提出了一种近邻密度分布优化的DPC算法.该算法在DPC算法搜索和发现样本的初始类簇中心的基础上,基于样本的密度分布采用两种样本类簇分配策略,依次将各样本分配到相应的类簇.理论分析和在经典人工数据集以及UCI真实数据集上的实验结果表明:文中提出的聚类算法能快速确定任意形状数据的类簇中心和有效地进行样本类簇分配;与DPC算法和KNN-DPC算法相比,文中算法在聚类效果与时间性能上有更好的平衡,聚类稳定性高,可适用于大规模数据集的自适应聚类分析.更多还原

【Abstract】 DPC algorithm is a new density based clustering algorithm that can automatically determine the number of clusters and cluster centers. However, there is a defect in the stability of clustering quality in the sample allocation strategy. KNN-DPC, an improved algorithm of DPC, has better clustering effect, but its practicality is affected by the low efficiency. In order to overcome the deficiencies of DPC algorithm and KNN-DPC algorithm, a neighbor density distribution optimized DPC clustering algorithm was proposed. Firstly, the algorithm searched and found the cluster centers with DPC algorithm. Then, two sample allocation strategies were adopted based on the neighbor density distribution of the sample, which was in turn used to assign the rest samples to the corresponding cluster. Theoretical analysis and the thorough experiments on several popular test cases include synthetic datasets and real-world datasets from UCI machine learning repository show that the clustering algorithm proposed can quickly determine the cluster center of arbitrary shape data and effectively perform sample cluster allocation. Compared with DPC algorithm and KNN-DPC algorithm, the proposed algorithm has a better balance between clustering effect and time performance, and has high stability. The algorithm proposed is an effective adaptive clustering algorithm that can be applied to large-scale data sets.更多还原