Improved density peaks clustering based on firefly algorithm
Jia Zhao, Jingjing Tang, Aiye Shi, Tanghuai Fan, Lizhong Xu
Abstract
The cut-off distance of the density peaks clustering (DPC) algorithm need to be set manually; the two local densities of the algorithm have a large difference in the clustering effect on the same dataset. To address the issue, the paper proposes an improved DPC based on firefly algorithm. It combines the cut-off kernel and the Gaussian kernel defined by the DPC algorithm, and balances the effects of the two kernels by the weighting factor. Meanwhile, a cluster-like centre evaluation criterion based on local density and relative distance of preference coefficient is constructed. In order to determine the parameters of the cut-off distance, weighting factor and preference coefficient, the three parameters are optimised by the firefly algorithm with the Rand index as the objective function. The experiment results show that the performance of the proposed method on synthetic datasets and real datasets is better than DPC and its variants.