Research On Adaptive Density Peak Clustering Algorithm

Density peak clustering algorithm(DPC)is a new clustering algorithm based on density.The algorithm has the advantages of simple principle,high efficiency and fast.Since it was proposed,it has attracted the attention of many scholars,and has been widely used in image processing,biomedical,document processing and other fields.At the same time,people also found some problems in the application of DPC algorithm:(1)the clustering results of the algorithm are affected by the parameter setting of the cutoff distance to some extent,and the parameter values set artificially can not avoid the subjectivity and randomness;(2)The calculation method of the local density of the sample only considers the distance factor,but fails to fully consider the global distribution;(3)It is necessary to manually select the clustering center after the generation of the decision graph.For some data sets,it may lead to multiple selection or missing selection of the clustering center,which will directly affect the clustering results.This paper conducts an in-depth study on the shortcomings of density peak clustering algorithm,and makes optimization and improvement on this basis.Specific research contents and results are as follows:1.An adaptive density peak clustering algorithm combined with whale optimization algorithm(WOA-DPC)is proposed.Firstly,a method of automatic selection of cluster centers is designed by using the weighted value of the product of local density and relative distance,which avoids the situation of less or more selection of cluster centers caused by manual selection.Secondly,considering that reasonable cutoff distance dc is an important factor to improve the clustering effect of DPC algorithm,an optimization problem with ACC index as the objective function was established,and the optimal cutoff distance dc was found by using the effective searching ability of whale optimization algorithm(WOA)to optimize the objective function.Finally,WOA-DPC algorithm is tested on synthetic datasets and real datasets.The results show that the three clustering index values obtained by WOA-DPC algorithm are all superior to other algorithms,and it has a good clustering performance on most datasets.2.A Density Peak Clustering Algorithm based on Weighted Shared Neighborhood and Accumulated Sequences(DPC-WSNN)is proposed.Firstly,the local density formula is redefined based on the weighted shared nearest neighbor,which not only avoids the influence of improper selection of cutoff distance dc on the clustering effect,but also can effectively deal with the uneven distribution of data sets of different classes.Secondly,on the basis of the original decision ? values,a set of ? cumulative sequences is generated,and the mean value of the cumulative sequences is taken as the critical point between the clustering center and the non-clustering center to realize the automatic selection of the clustering center.Finally,the DPC-WSNN algorithm is tested on 16 datasets,and the results show that the DPC-WSNN algorithm has better clustering effect than other comparison algorithms.