Research Of Fuzzy Clustering Based On Artificial Fish Swarm Algorithm And Its Application On Wine Categories

Compared with traditional methods of hard clustering, fuzzy clustering algorithm candescribe the uncertain relation of data accurately and has a better ability to express data whichmakes it become the mainstream of the clustering algorithms in recent years. The objectivefunction based fuzzy clustering algorithms are the most popular used method in fuzzyclustering.The representative algorithms of fuzzy clustering algorithms are the KFCM andFCM algorithm.This dissertation mainly focus on the studying the FCM algorithm and KFCMalgorithm, then some improved algorithms are proposed to overcome the defect of the originalalgorithms. The improved algorithms are applied to the wine categories at last.FCM algorithm is particularly sensitive to initialization, and easy to fall into localminimum values，so it only can get a local optimal solution rather than a global optimalsolution. Artificial fish swarm algorithm is an autonomous body optimization mode based onanimal behavior. It is derived from the foraging behavior of fish. AFSA is not sensitive to thechoice of the initial values and parameters, has a good ability to obtain the global extremumand easy to implement. So this dissertation introduced the AFSA into the FCM algorithm,Experimental results show that the new algorithm(AFSA-FCM) solved the defects of FCMwhich is sensitive to the initial value and easy to fall into the shortcomings of local minima.KFCM algorithm integrates FCM algorithm with the kernel technique to construct akernel clustering algorithm.By using Mercer kernel functions,the data can be mapped in theoriginal space to a high-dimensional feature space in which the accuracy clustering result canbe improved efficiently compared with FCM algorithm. The reason is that KFCM algorithmovercomes the reliance of the inner shape of the distribution of the data to a certain extent.Butthe KFCM algorithm is also sensitive to initialization, and easy to fall into local minimumvalue，to solve these problems, this dissertation introduced the AFSA into the KFCMalgorithm(AFSA-KFCM).The iteration process is replaced by the AFSA based on the gradientdescent of KFCM. AFSA-KFCM have a strong global searching capacity and avoid the localminimum problem of KFCM. Experiments show that the proposed algorithm is more accurateand efficient than KFCM.Fuzzy clustering is used to classify the quality of wine in this dissertation.Firstly, thebest clustering number of wine dataset is determined by the clustering validity functions,andthen the fuzzy clustering algorithms which are proposed by this dissertation are applied to theclassification of wine. Experiment results show that the improved fuzzy clustering algorithm can improve the wine classification accuracy. to some extent.