Research On Algorithms Of Fuzzy Support Vector Machine

Support vector machine(SVM),which was proposed in 1990s by Vapnik etc,is one of the standard tools for machine learning and data mining. It can deal withclassification problems and regression problems successfully. Because of itsexcellent learning performance, this technology has been the hot topic of machinelearning. But as a new technology, it still has some problems. There are lots offuzzy information in the objective world, and if the training set has fuzzyinformation(fuzzy parameters) in SVM, traditional SVM will become poor andindeed will fail.This dissertation proposed an improved Fuzzy Support VectorMachine(FSVM) algorithm to solve the shortage when the training set has fuzzyinformation in the samples. First, the structure theory and basic theory-statistical learning theory are analysed and researched. Furthermore, the modifiedthe FSVM algorithm is proposed. In the model of modified FSVM, the innerconstructing procedure and the exterior structure of FSVM are describedmathematic approach and using the FSVM as classifier classification andrecognition for extracted feature are carried out. Finally, the experimental data ofthis model with that of the other SVMs are compared.The results of experiment show that the proposed FSVM can preferablyresolve the classification problem for the samples with fuzzy information,comparing with some other SVMs algorithm this algorithm has excellence ofreducing the sample space size and the requirement of memory, and the speed oftraining and classification is improved.