Researching Of Mutiple Classification Method For Support Vector Machine Expansion

Support vector machines(SVMs)are new developed machine learning methods based on the statistical learning theory(SLT).The main advantage of SVMs is that it can serve better in the processing of small samples by substituting structural risk minimization(SRM)for empirical risk minimization(ERM).Moreover,SVMs can solve practical problems such as over learning,high dimension and local minima,which exist in most of learning methods.Currently,being the optional learning theoryof small samples,SVMs are attracting more and more researchers and becoming a hotspot in the fields of artificial intelligent and machine learning.SVM was designed for 2-type classification t first,but most of the time people need multiple classification.So it needs expanded method to reach real world use,now there are two main methods,one kind is to modify classification directly function to combine multiple problem into one optimization problems,this way seems simple,but actually it has a high computation complexity and can`t be achieve easily,though there are a few studies exist,the essential problem still can`t be solved.Another way is to combine classifiers to construct multiple classifier,it`s a hotspot in SVM researching field.The main work of this thesis is as follows.1.This paper introduces basic theory and research status of the statistical learn theory and machine learning,gives introduction and comparison.Discussed the base theory of SVM and how the problem is solved.2.Summary and comparison of the common multi-class classification method are given,and for DAG-SVM,gives out a high accuracy DAG structure choosing method that using the classificatory percentage in training data set to estimate the accuracy of the test data set,and then to choose the one got highest accuracy,which was supposed to reach a better use in the real world.The experiment shows the structure has a better effect in chosen testing data set,the method has a nice prospect in data set which need high accuracy.