Research On The Technology Of Creating And Optimizing High-dimensional Fuzzy Classifier Based On Genetic Algorithm

At the information-based ages, the data is sufficient, while the information is insufficient. And under this circumstance, Data mining comes out. It is a process which extracts the hiding and unknown potentially useful information and knowledge from large amounts of incompletely noisy obscure random datas. Classification is learnt to use a classifying function or classifying model-classifier, which reflects the data in the database to a class in the given class. It can also used to pick-up the data model or forecast the trend of the future data.The future study on the method of the data classification will focus more in intelligent classification area, such as the study of Ant Colony Optimization,Genetic Algorithm,Paticle Swarm Optimization and other classification, as well as using hybrid algorithm to classify. This paper is based on the improvement of the GA,and designs and creates the Mamdani Fuzzy Classifier based on the Mamdani fuzzy logical system by using the improved GA,and obtains the better classifier finally.The first step of this paper is the improvement of GA,viz. a hybrid genetic algorithm based on clan competition. Basic GA and Evolutionary Programming belong to the evolutional algorithm theory. They have a good optimization for real-valued-continual function, and their convergent probabilities are 1 in theory, but they are difficult to converge at global optimal solutions in practice. Through the analysis and experiment of these two algorithms, and according to the idea of clan competition, a new hybrid genetic algorithm based on clan competition is proposed. It is proved that the probability of the new algorithm convergent to the global optimal solution is 1, when compared with the former two algorithms. They are applied to typical function optimization problems and experiments results illustrate that this new algorithm is the robustest among the three algorithms. What's more, it has the highest computation precision with the equal parameters. The second step of this paper is to restructure the Mamdani fuzzy logical system to form the Mamdani fuzzy classifier. Fuzzy system is a fuzzy mapping to a given input area,the result of this mapping is a limited output area. Mamdani fuzzy system's output is a continual area,then some classes is formed by proper partitioned the area, thus, the fuzzy classifier is created。The third step of this paper is to use the new GA to select and optimize the Mamdani fuzzy classifier. Let alone the output area's dimension, separates them in coding and selects and classify the variable to form precondition of rule, and gets the postcondition through practicing. Then, a fuzzy rule set and a fuzzy classifier is formed. Taking accuracy and interpretation as target function, the Mamdani fuzzy classifier is optimized.In order to prove the validity of new GA and the improved Mamdani fuzzy classifier, selected the classical optimized function and Iris data-base is selected as examples respectively to simulate. Comared with the result of the related documents, the new GA and Mamdani fuzzy classifier works well.