A dynamic shuffled differential evolution algorithm for data clustering

In order to further improve the convergence performance of data clustering algorithms, a dynamic shuffled differential evolution algorithm, DSDE for short, is presented in this paper. In DSDE, mutation strategy DE/best/1 is employed, which can take advantage of the direction guidance information of best individual so as to speed up the corresponding algorithm. Meanwhile, inspired by shuffled frog leaping algorithm, a sorting scheme and a randomly shuffled scheme are used to divide a total population into two subpopulations during the evolving process. In this way, mutation strategy DE/best/1 is actually used in two subpopulations, respectively, which can effectively exchange information between two subpopulations and balance the exploitation ability of DE/best/1/bin. In addition, most popular data clustering algorithms suffer from the choice of initial clustering centers, which may cause a premature convergence. Here a novel initial technique, called the random multi-step sampling, is integrated into DSDE to overcome the shortcoming. Then an experiment tested on 11 well-known datasets has been carried out, and the related results demonstrate that DSDE significantly outperforms DE/rand/1/bin and DE/best/1/bin. Next, another comparison among DSDE and other four well-known data clustering algorithms is conducted. The related results also show that DSDE is superior to other four approaches including particle swarm optimization with age-group topology (PSOAG) in terms of objective function value, i.e., the sum of intra-cluster distance. In a word, all the experimental results confirm that the proposed algorithm DSDE can be considered as an excellent tool for data clustering.