Peak density algorithm based on KD-tree optimization

Abstract

In view of the high time complexity of the density peak algorithm, it is necessary to manually confirm the clustering center according to the decision graph. A density peaking algorithm KT-DPC based on kd-tree optimization is proposed. The algorithm defines the local density p through K-nearest neighbor and uses kd-tree to accelerate the local density p and distance 8. In addition, in the confirmation stage of clustering center, a clustering center confirmation strategy (C2BD, clustering center confirmation based on difference) is proposed. The difference between adjacent y is calculated by arranging y=p*S in ascending order, and the boundary between clustering center and non-clustering center is found according to the change of difference. This method automatically confirms the clustering center of the algorithm, avoiding the problems of strong subjectivity and insufficient clustering accuracy caused by manually confirming the clustering center. Experiments on multiple UCI public data sets show that the running time and clustering accuracy of the KT-DPC algorithm under low-dimensional data are better than traditional DPC algorithms, KNN-DPC and other improved algorithms.