An improved density peak clustering algorithm

Abstract

As the density peak clustering algorithm is insensitive to local structure, having bad clustering effect on data sets with uneven density distribution, and the algorithm does not take into account the hierarchy of data points, an improvement of density peak clustering algorithm is presented. The contribution factor is introduced into the local density calculation of data points to improve the insensitivity of the original algorithm to the local structure. In addition, the concept of trend is adopted to reduce the allocation joint error caused by the original algorithm only judging the similarity according to the distance of data points, and the method of leading forest is proposed to explain the hierarchy of data points. Finally, the experiments on artificial data sets and UCI datasets show that the improved algorithm is feasible.