基于密度峰值及肘方法的类数目确定

Determination of the number of classes based on density peak and elbow method

为确定数据集真实类数目,降低参数对于密度峰值算法的影响,本文基于自然邻居思想提出了密度峰值(DP)算法的新内核,以摆脱参数的限制,实现真正的自适应聚类。将密度峰值算法与肘方法联合,提出了自然密度峰(NDP)指标。此外,通过减而治之的思想,改进了自然邻居的搜索算法。新的指标和方法可以适用于多种类型的数据集(例如线形、流形、环形和凸形数据集)求得最优类簇数目。经理论研究和实验结果分析可以证实所提出的指标和算法的有效性。

In order to determine the number of real classes in the data set,and reduce the influence of parameters on the density peak(DP)algorithm,this paper first proposes a new kernel of the density peak algorithm based on the idea of natural neighbors to get rid of the limitation of parameters and realize true adaptive clustering.Then,a new clustering effectiveness index of natural density peak(NDP)is proposed by combining the density peak algorithm with the elbow method.In addition,the search algorithm for natural neighbors is improved through the idea of reducing and conquering.The new indicators and methods can be applied to multiple types of data sets,such as linear,manifold,circular and convex data sets,to find the optimal number of clusters.The theoretical research and analysis of the experimental results can prove the effectiveness of the proposed index and algorithm.