基于子簇融合和线性判别分析的密度峰值聚类算法

DPC algorithm based on sub-cluster fusion and LDA

密度峰值聚类(DPC)算法有能够发现非球形簇等优点。但在算法中,局部密度和最近邻距离计算易忽略样本间相关性,并且算法在高维数据集上聚类效果较差。针对上述问题,提出一种基于子簇融合和线性判别分析的DPC算法(SCF-LDA-DPC)。首先,引入样本间Pearson相关系数构造加权高斯核密度估计函数计算局部密度。其次,设计一种子簇融合策略,避免数据错误分配,优化算法容错性差缺陷。最后,引入LDA算法对高维数据降维,提高DPC算法鲁棒性和准确性。多个数据集实验结果表明:SCF-LDA-DPC算法在聚类精度和聚类性能方面明显优于其他优秀算法。

Density peak clustering(DPC)algorithm can find non-spherical clusters and other advantages.However, local density and the nearest neighbor distance computation are easy to ignore the correlation between samples, and the clustering effect of the algorithm is poor on high-dimensional dataset.To solve these problems, a DPC algorithm based on sub-cluster fusion and linear discriminant analysis(SCF-LDA-DPC) is proposed.Firstly, the weighted Gaussian kernel density estimation function is constructed by introducing Pearson correlation coefficient between samples to calculate local density.Secondly, a seed cluster fusion strategy is designed to avoid data misallocation and optimize fault tolerance of algorithm.Finally, LDA algorithm is introduced to reduce the dimension of high-dimensional data to improve the robustness and accuracy of DPC algorithm.The experimental results of multiple datasets show that the SCF-LDA-DPC algorithm is superior to other excellent algorithms in clustering precision and performance.