A multi-clustering method based on evolutionary multiobjective optimization with grid decomposition

Abstract At present, it is a challenging task to determine the number of clusters (k), which has a great impact on the quality of major clustering methods. Multi-objective evolutionary algorithms (MOEAs), which determines k value adaptively, have been widely adopted for clustering. However, when the range of k becomes increasingly large, Pareto Front (PF) approximations obtained by an MOEA may not be uniformly distributed, leading to the difficulty of obtaining the optimal k value for clustering. For this reason, an MOEA base on constrained decomposition with grids (CCDG-K) is designed for clustering and better identify the optimal k value. CCDG-K adopts a grid system for decomposition. The grid system has an inherent property of reserving diversely populated solutions, which is very desirable for clustering. The experimental studies show that CCDG-K can deliver solutions of all k values which is of great help for obtaining the optimal k value. The experimental results also indicate that CCDG-K outperforms other algorithms in clustering.