An Evolutionary Multi-centers Based Dynamical Clustering Algorithm

Clustering method is one of the important data mining techniques which is used to group objects into a pre-specified number of clusters. However to many datasets it is usually hard for users to estimate correctly its number of clusters. To solve the problem of existing clustering algorithms, a multi- center based evolutionary clustering method is proposed in this paper. To dynamically determine the number of clusters well and robustly on irregularly structured datasets, multiple centers are assigned to each cluster and each individual is evaluated with two modified fitness functions, i.e. heuristic intra-cluster variation fitness function and connectivity fitness function. The corresponding evolutionary operators are designed. Experimental results on the UCI datasets and artificial datasets show that our proposed method can obtain good clustering results and outperforms other comparative methods.