Research Of Improved Artificial Bee Colony For Fuzzy C-means Clustering Algorithm

With the development of computer technology, different kinds of information are explosive growth, meanwhile computing and storage capacity are also rising. So how to extract useful information from the complex data to help us making analysis and decision is getting more and more attention. As an important branch in the field of data mining, clustering for large amounts of data plays an irreplaceable role. The data in the form of diversification and abundant, also take more and more strict in clustering. Traditional clustering algorithm is sensitive to the initial point, poor ability of divided, and more and more can’t satisfy people’s needs.Artificial bee colony algorithm is not sensitive to the initial point and has the advantage in searching ability and adaptability. To solve the problem that artificial bee colony algorithm in one-peak problems convergence speed is slow and multi-peak problems easy to fall into local optimum faults, the mutation and crossover thought of differential evolution algorithm are introduced to improve swarm algorithm convergence speed and balance the global and local search ability.Fuzzy c-means clustering algorithm has a wide range of usage in data mining. Due to sensitivity to initial point and poor search ability, it limits the further application of the algorithm. Combining the improved artificial bee colony algorithm with the fuzzy C-means clustering algorithm, and running in a number of international standard data sets for experimental verification, the results show that the differential evolution artificial bee colony based fuzzy C-means algorithm have been achieved satisfactory results.In order to further promote the differential evolution artificial bee colony based fuzzy C-means algorithm. We have done a large number of experiments, and summarize the algorithm’s rule of two important control parameters. Increasing the value of mutation factor (F), can lead to the increase of diversity of population, then lower the risk of early reduction, stability increased, but the algorithm’s convergence speed will be decline. When the CR value increasing, can accelerate the convergence speed and reduce the number of clustering iteration. To the contrary, it will improve the global search ability and the rate of convergence. But we can’t increase the value of CR blindly, because when the CR value beyond a certain threshold, will lead the algorithm to stochastic searching with rising convergence speed. Moreover the CR values need to be determined according to the actual data sets, too large will reduce ability of global searching.For complex clustering, it is easier to lose the global optimal solution.