Currently, Fuzzy C means algorithm (FCM) is a popular and widely used fuzzy clustering algorithm. It has been successfully applied to pattern recognition, image processing and other fields. However, it requires the existence of prior knowledge, sensitive to initial value, and easy to fall into local minima and other inherent weaknesses.To compensate for the shortcomings of FCM algorithm, and in view of excellent optimized results of evolutionary algorithms as a stochastic global search technology, we combined genetic algorithm with differential evolution algorithm proposed a heterogeneous coevolutionary clustering algorithm - GADEFCM algorithm. The algorithm can automatic decide the number of clusters via using an improved masker manner. Dividing population into two subpopulations that constitute of same size individuals, in which a subpopulation uses improved genetic algorithm to evolve and the other subpopulation uses difference evolutionary algorithm with hybrid mutation strategy which is proposed in this paper to evolve. Then, in purpose of accelerate algorithm convergence speed, every individual in the population implement FCM algorithm according to rules. In the process of evolution, two subpopulations exchange best individual to each other through heterogeneous interval migration policy to guide the search, which can balance the exploitation ability and the exploration ability of our algorithm. GADEFCM algorithm overcomes the shortages of FCM algorithm of requiring prior knowledge of the number of clusters and easy getting into local minimum, it can be more efficient to find the global optimal solution.In this paper, we make experiments with MATLAB simulation platform, which using data sets for tests of clustering results, the number of clusters and run time on GADEFCM algorithm and the other algorithms. The experimental results show that the algorithm can find the correct number of cluster centers, obtain better clustering accuracy on the premise that run time is allowed. Finally, GADEFCM algorithm is applied to text clustering and the experiments proved that the algorithm can obtain better clustering results. |