On the Adaptive Selection of the Cutoff Distance and Cluster Centers Based on CFSFDP

Aiming at deficiencies of clustering by fast search and find of density peaks (CFSFDP) algorithm in the manual selection of the cutoff distance and cluster centers, this paper proposes an adaptive CFSFDP (A-CFSFDP) algorithm, which is based on kernel density estimation and anomaly detection. Firstly, the distribution characteristics of data points are analyzed by using non-parametric kernel density estimation and the optimal value of the cutoff distance is adaptively selected by iteration. Secondly, the idea of anomaly detection is adopted and by calculating the weight of the cluster center for each point, anomaly points can be detected as cluster centers of each cluster. Finally, the idea of local density within the cluster is introduced and by computing it for each data point, the optimized division of cluster core and cluster halo can be achieved. The simulation results show that the proposed algorithm can not only avoid the subjective factors in the selection of cutoff distance and cluster centers compared to CFSFDP, but achieve better performances in robustness and accuracy of clustering.