A fuzzy density peaks clustering algorithm based on improved DNA genetic algorithm and K-nearest neighbors

Abstract

In recent times, a density peaks based clustering algorithm (DPC) that published in Science was proposed in June 2014. By using a decision graph and finding out cluster centers from the graph can quickly get the clustering results, easy and efficient. While, in terms of local density measurement, DPC does not adopt uniform density metrics. Instead, uses different local density metrics according to the dataset size. In addition, when the size is small, the subjective choice of the cutoff distance dc has a greater impact on the clustering results. In order to make up for the defects of DPC and improve the performance of this algorithm, we propose a fuzzy density peaks clustering algorithm based on improved DNA genetic algorithm and K-nearest neighbors (named as FDPC+IDNA). On one hand, FDPC+IDNA uses fuzzy neighborhood relation to unify the local density metric which combines the high efficiency of DPC algorithm with the robustness of fuzzy theory. On the other hand, we introduce the idea of K-nearest neighbors and an improved DNA genetic algorithm to compute the global parameter dc that improves the shortcomings of empirical judgment. Experiments on synthetic and real-world datasets demonstrate that the proposed clustering algorithm outperforms DPC, DBSCAN and K-Means.