Researches And Applications In Fuzzy Cluster Technology Based On Genetic Algorithms

Among the clustering algorithms basing on the objective function, the theory of fuzzy c means algorithm (FCM) is the most perfect and its application is widest. However, it is essentially a local search algorithm and finds the optimal solution by using the iterative hill-climbing technique. Therefore, it has a fatal weakness which is very sensitive to initialization and gets into the local minimum value easily. Genetic algorithm is a global optimization method and used extensively. Its main advantages are simple, universal, robust, and more efficient than objectless searching and more popular than the algorithm which aims at a specific problem, and which is unrelated to the model of solving the question. These characteristics of genetic algorithms overcome the weakness of FCM. Combining genetic algorithm with the FCM not only plays the genetic algorithm ability in global optimization, but also takes into account the capacity of local optimization of FCM. Hence it can improve the convergence rate and solve the clustering problem better.In this paper, an improved genetic fuzzy clustering algorithm (IGFCM) is presented on the basis of FCM algorithm, genetic algorithm and genetic clustering algorithm, which uses the genetic algorithm to optimize the initial cluster centers, and then carry out the FCM algorithm. Genetic algorithm combines with FCM to make up for the deficiencies of the single one of them, and increases the convergence rate and makes better classification results. First of all, in the genetic algorithm, the symbol encoding mode of the cluster centers is used as the chromosome. It not only shortens the length of the chromosome, but also maintains the search space unchanged after the crossover and mutation operations. Secondly, the optimum preservation strategy is adopted in the selection operation to maintain the optimum individual in the process of genetic, and then the selected individual enters the next generation directly without participating in crossover and mutation operation, and further then the roulette method is employed to pick out the suitable individual according to the corresponding probability distribution. Simultaneously, crossover and mutation operated to improve the average fitness of groups and ensure the current optimal individual will not be damaged by genetic manipulation in the evolutionary process of each generation. Finally, the maximum number of iterations and in accordance with the degree of genetic convergence criteria are set as stopping criterias, which reduce the errors and shorten the run-time of genetic algorithm.In this theses, numeric simulations are carried out by MATLAB, the FCM algorithm, GFCM algorithm and IGFCM algorithm are tested by IRIS data sets, respectively. Furthermore, the the sensitivity of the FCM algorithm can be overcomed by the IGFCM algorithm is verified, the more effective of IGFCM algorithm in run-time than the GFCM algorithm is proved, and at last the application of clustering algorithm in text clustering is discussed.