Researches In Kernel-based Fuzzy C-Means Clustering Algorithm Based On GA Optimization

With growth of data resource, data mining has become an indispensible method of dataanalysis to get useful information. As an available tool of data mining, unsupervised fuzzyclustering analysis has been applied efficiently in many fields such as information retrieval,pattern recognition, data analysis and image processing. The research is focused on theoptimization and application of the fuzzy clustering algorithm in this paper.For the fuzzy clustering algorithms, the fuzzy c-means (FCM) clustering algorithm based onthe objective function is widely applied because of its strong ability of local search and its fastconvergence speed. However, FCM algorithm has two defects. First, to one sample, the sum ofthe membership degree for all categorizations is equal to1, which makes it sensitive to the noiseand isolated data. Second, FCM is essentially a kind of local hill-climbing algorithm, whichmakes it sensitive to the initial clustering center and easy to converge to a local extremum.Aimed at the problems existed in the FCM clustering algorithm, a kernel-based fuzzyc-means clustering algorithm is proposed to optimize fuzzy c-means clustering, based on theadaptive GA optimization (GA-KFCM) which is combined of the improved genetic algorithmand the kernel technique. Firstly, an improved adaptive genetic algorithm is designed by the realcoding mode, non-linear ranking select measurement, adaptive crossover and mutation strategy.The criteria are the maximum evolution iteration and the average fitness convergence. Then, akernel-based FCM clustering algorithm (KFCM) is presented by change the clustering distanceof the FCM to define the objective function, and then to improve the constraint conditions ofprobability in FCM. Finally, an algorithm called as GA-KFCM is proposed, which is combinedthe improved adaptive genetic algorithms presented in this thesis with the KFCM clusteringalgorithm. In this algorithm, the improved adaptive genetic algorithm is used to optimize theinitial clustering center firstly, and then the KFCM algorithm is availed to guide thecategorization, so as to improve the clustering performance of the FCM algorithm.In this thesis, Matlab is used to realize the simulation, and the performance of FCMalgorithm, KFCM algorithm and GA-KFCM algorithm are testified by IRIS and WINE datasets,respectively. The results proved that the GA-KFCM algorithm proposed in this thesis overcomesFCM’s defects efficiently and improves the clustering performance greatly. Based on the results,the GA-KFCM is applied in text categorization, and its performance is testified effective.