Research On Fuzzy Support Vector Machine Algorithm Based On Structural Information

Support vector machine is a machine learning method based on the VC dimension theory and the principle of structural risk minimization of statistical learning.This method shows many unique advantages in dealing with classification problems,and it has been widely used in many fields.However,as a traditional method,support vector machine still has many limitations.Therefore,researchers have made a thorough study of its existing problems and proposed a variety of improved algorithms,such as fuzzy support vector machine,structured regularized support vector machine,support vector machine with parametric margin and so on.In order to further improve the performance of the algorithm,we study fuzzy support vector machine and propose fuzzy support vector machine based on structural information and structural fuzzy support vector machine with parametric margin.The main research contents include:1.By introducing structural information into fuzzy support vector machine,we propose a fuzzy support vector machine based on structural information.Although the fuzzy support vector machine applies fuzzy theory to support vector machine,the effect of different samples on the production of optimal classification is different by degree of fuzzy membership,and the effect of noises or outliers on support vector machine is better solved,the algorithm mainly takes advantage of the separability between classes of data sets and ignores the role of structural information within the dataset classes for classification.Therefore,on the basis of the study of the fuzzy support vector machine and the structural regularized support vector machine,the structural information is introduced into the fuzzy support vector machine,we propose a fuzzy support vector machine based on structural information.The method not only uses degree of fuzzy membership to show the effect of different samples to support vector machine.but also improves the generalization performance of classifier by using the structural information within the classes.2.The parametric margin is introduced into the fuzzy support vector machine based on structural information,and a structural fuzzy support vector machine with parametric margin is proposed.Due to the heterogeneous noise existing in the data set,it is easy to make the classification edges appear irregular shape.Based on the structural fuzzy support vector machine,by considering the influence of the heterogeneous noise on the model,we propose a structural fuzzy support vector machine with parametric margin model.The model further improves the classification accuracy of the support vector machine.3.For the proposed the fuzzy support vector machine based on structural information and structural fuzzy support vector machine with parametric margin,we use the Lagrange multiplier method to theoretically derive the classifiers of the fuzzy support vector machine based on structural information,the v-fuzzy support vector machine based on structural information and the structural fuzzy support vector machine with the parametric margin.At the same time,we introduce the nuclear strategy into the proposed models and solve them to solve the classification of more complex data.4.By selecting the standard data sets in UCI and Statlog database and obtaining the structural information by calculating the within class scatter,the performance of the fuzzy support vector machine based on structural information and structural fuzzy support vector machine with parametric margin is studied experimentally.The performance of the methods is also compared with the performance of the methods such as support vector machine,fuzzy support vector machine,structure regularized support vector machine and support vector machines with parametric margin.