An Elitist-Based Differential Evolution Algorithm for Multiobjective Clustering
Mingzhu Zhang, Jie Cao · 2020
In this paper, we convert the clustering problem with an unknown number of clusters into a multiobjective optimization problem, and propose a novel elitist-based differential evolution algorithm for multiobjective clustering (EDEMC). It aims to minimize the number of clusters and maximize the compactness within clusters simultaneously, and generates a Pareto-optimal set consisted of multiple clustering solutions for different cluster numbers. These two optimization objectives are essential factors for clustering. EDEMC creates and maintains an elitist archive which stores historical best solutions for each number of cluster, and it iteratively optimizes the population with newly designed genetic operations and replenishment strategy. In the end, users could flexibly choose one optimal partitioning of a certain number of clusters by some preferred criteria from the solution set. Experimental results on several datasets illustrate that the proposed method can provide more convergent and diverse solutions in a shorter time.