The Model Research On Fuzzy Classification

Fuzzy classification is an important application of fuzzy set theory.Fuzzy classification rules are widely considered a well-suited representation of classification knowledge. Since they resemble the way which humans would possibly formulate their knowledge,they are readable and interpretable. Fuzzy classification has been widely applied in many fields, such as image processing, words recognition, voice recognition, text classification, remote sensing, weather and industry automation.The key problems of the model research on fuzzy classification are the automatic generation of fuzzy partitions and fuzzy classification rules, the expression of fuzzy rules, the optimization tuning of classification rules and the improvement of classification recognition rate.To study above problems, three fuzzy classification models and a method of classifier ensemble are proposed with different views, different methods and techniques in this paper.Model I: Fuzzy classification model basing fuzzy kernel hyperball perceptron (FCMBFKP). In this model, firstly the patterns in the initial input space are mapped to high dimensional feature space by selecting a suitable kernel function.In the feature space, the hyperball which covers all training patterns of a class is founded for every class by the algorithm of fuzzy kernel hyperball perceptron. a hyperball is regarded as a fuzzy partition and a IF-THEN rule is created for a fuzzy partition.A hyper-cone membership function is defined with regarding the center and radius of the hyperball as parameters.Considering the possibility of overlapping areas among hyperballs, the policy and algorithm of tuning the rules are proposed with regarding the radius of the hyperball as tuning parameter. Experiments with the data sets of standard machine leaning database evaluate the performances of this model with comprison of experiment results of the methods of kernel hyperball perceptron (KHP) and SVM.Model II: Fuzzy classification model basing evolving kernel clustering (FCMBEKC). Similar to the model I, firstly the patterns in the initial input space are mapped to high dimensional feature space by selecting a suitable kernel function. In feature space, Training patterns of every class are clustered by the proposed evolving kernel clustering algorithm.For each class, multiple hyperballs are got and a hyperball is regarded as a cluster which relates to a fuzzy partition. An IF-THEN...