Font Size: a A A

The Optimization Of SVM On Parameter Selection

Posted on:2009-08-07Degree:MasterType:Thesis
Country:ChinaCandidate:J WangFull Text:PDF
GTID:2178360245456722Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
The support vector machine(SVM),which is based on the statistical learning theory(SLT)and the structural risk minimum principle,guarantees the largest generalization ability of a model.It is,therefore,theoretically more perfect than the neural network model that is based on the empirical risk minimum principle.SVM is a nonlinear classifier with multi-classifcation performances.In solving classify problem,such as trouble examine,if there is a big number discrepancy of samples in different classes,C-SVM is undesirably biased towards the class with fewer samples,so,the training accuracy is unsatisfied.In order to improve accuracy,an optimizing algorithm is proposed based on the different weight with different classes in the training course in this paper.According to positive and negative proportion of samples in the total samples,the weight of samples toward to the class of fewer number samples is increased;the other is decreased,realizing the balance between two samples.It is showed with experiments that the proposed approach can improve the accuracy rates of classification.The paper also established the pattern recognition classifiermodel and studied the parameters that influence the classifier model's classification ability;on the basis of analyzing the parameter's influence on the classifier's recognition accuracy,it proposed the selfadaptive optimization algorithm for the SVM classifier model using genetic algorithm.After that,calculation instances show the effectiveness of the optimization algorithm.
Keywords/Search Tags:SVM, pattern recognition, genetic algorithm, optimization
PDF Full Text Request
Related items