An effective hybrid method based on DE, GA, and k-means for data clustering

Abstract

Clustering is an unsupervised classification method and plays essential role in applications in diverse fields. The evolutionary methods attracted attention and gained popularity among the data mining researchers for clustering due to their expedient implementation, parallel nature, ability to search global optima, and other advantages over conventional methods. However, conventional clustering methods, e.g., K-means, are computationally efficient and widely used local search methods. Therefore, many researchers paid attention to hybrid algorithms. However, most of the algorithms lag in proper balancing of exploration and exploitation of solutions in the search space. In this work, the authors propose a hybrid method DKGK. It uses DE to diversify candidate solutions in the search space. The obtained solutions are refined by K-means. Further, GA with heuristic crossover operator is applied for fast convergence of solutions and the obtained solutions are further refined by K-means. This is why proposed method is called DKGK. Performance of the proposed method is compared to that of Deferential Evolution (DE), genetic algorithm (GA), a hybrid of DE and K-means (DEKM), and a hybrid of GA and K-Means (GAKM) based on the sum of intra-cluster distances. The results obtained on three real and two synthetic datasets are very encouraging as the proposed method DKGK outperforms all the competing methods.