自适应截断距离与样本分配的密度峰值聚类算法

Density Peak Clustering Algorithm Based on Adaptive Cutoff Distance and Sample Allocation

针对密度峰值聚类算法中,样本局部密度截断距离需主观选择和样本分配策略的误差扩散问题,提出自适应截断距离和构造流形距离优化样本分配的改进型密度峰值聚类算法。该算法首先使用样本K近邻自适应的选取各点的截断距离,即在样本密度大的点,选用大截断距离,准确选取类簇中心,在样本密度小的点,选用小截断距离,判别离群点。其次对于剩余样本通过样本的连接路径构造流形距离,优化样本分配策略。最后选取人工数据集进行聚类分析算法实验,与传统的密度峰值聚类算法进行实验对比,验证所提改进算法对聚类中心选取和样本分配的准确性。

Aiming at the shortcomings of the subjective selection of the cutoff distance and the sample allocation strategy in the sample local density of the density peaks fast search clustering algorithm,an improved density peaks clustering algorithm which is adaptive to the cutoff distance and Manifold distance optimization is proposed.The algorithm uses the sample K nearest neighbor adaptive selection cutoff distance.In the place where the sample density is large,the large cutoff distance is selected to accurately select the cluster center.For the remaining samples,an optimized sample allocation strategy of manifold distance was adopted.Artificial data sets were selected for clustering analysis in the algorithm verification experiment,and the experiment was compared with the traditional peak density clustering algorithm to verify the accuracy of the improved algorithm in clustering center selection and sample allocation.