Classification Algorithm Based On Support Vector Machine

The traditional pattern recognition methods deal with the issue with the accurate model. It is also difficult to have a good showing against the non-linear problem. The traditional statistical learning theory is asymptotic theory based on the number of the sample tending to infinity. Practical applications of these methods are often unsatisfactory. The support vector machines have good displays in solving modeling and the dimension disaster. It has also the good application in the small samples. It is becoming a new hot spot in the machine learning domain.This thesis discusses the principle of the support vector machine and its application in classification. The experiments are divided into three groups and two kinds of support vector machines for polynomial kernel and radial basis function are chosen to experiment by changing the parameter values.In the classification experiment, we find that the number of the support vector is far less than the number of the training sample number. This provides the method for us to solve the large-scale data problem by dividing train samples into several small subsets and sequentially training subsets one by one. Different size training samples are chosen in the experiments. The results show that, based on structural risk minimization principle, the less training samples, using simple structure learning function, will be to avoid a over fitting phenomenon. Contrary to more training samples, slightly simple structure of the learning function will reduce promotional ability. In the experiment, we introduce penalty factor C to allow the training samples to be wrong classified. Appropriately increasing the value of the penalty factorC , the promotional ability of the learning machine can be improved. The support vector machine has avoided the dimension disaster with the inner product operation. The parameters of the support vector machine would affect the classification results in the medium-scale training samples. In order to get the best classification accuracy, we are using cross-validation method to choose parameter values. The results show that cross-validation method increases the average accuracy. Finally, we compare the results of the support vector machine algorithm and the neural network algorithm. The results show that the support vector machine algorithm is superior to the neural network algorithm.